JP2011053058A - Vehicle inspection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a vehicle inspection device including a reference mark detection technique for, in the detection of the feature amount of a scratch on the side face of a wheel or the like, specifying the position of the scratch, or the like, based on the relationship to the absolute position. <P>SOLUTION: The vehicle inspection device detects a feature amount on a wheel from an input image of a side face of a wheel for railway vehicles, detects the coordinate of the center of the wheel from the input image of the side face of the wheel for railway vehicles; detects a reference mark from the input image of the side face of the wheel for railway vehicles, determines the coordinate thereof; and stores the position of the feature amount determined using the coordinate of the wheel center and the coordinate of the reference mark as references. Since the reference mark which is the absolute position for specifying the position of the feature amount can be detected, it is possible to automatically inspect a vehicle. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a vehicle inspection apparatus that automatically detects a feature amount of a side surface of a wheel by image processing.

  In the inspection of railway wheels, conventionally, visual observation has detected feature quantities such as scratches and judged deterioration. However, the visual inspection can be performed only on the visible portion of the stopped wheel, and in order to inspect the entire circumference of the wheel, it is necessary to remove the wheel for inspection. For this reason, it has been desired to automate inspections using images taken during traveling.

  On the other hand, by grasping the growth history of features, there is a need to utilize it for preventive maintenance applications such as life prediction and collation with driving conditions, but for history management, the same wheel is used between different inspections. It is necessary to check the coordinate system.

  In order to satisfy these needs, it is necessary to detect a dynamically changing reference position from an image taken during traveling.

  Japanese Patent Application Laid-Open No. 7-14672 is known as a prior art relating to automation of wheel inspection. In this patent document, a camera is installed beside a track, an image of a wheel tread is acquired based on a signal from a sensor that detects the passage of a wheel, and a feature amount is detected using image processing.

Japanese Patent Laid-Open No. 7-174672

In the prior art disclosed in the above patent document, the feature amount of the wheel tread is detected. However, the present invention intends to automatically detect the feature amount generated on the wheel side surface by image processing.
Further, in the prior art, since there is no means for collating the positions of flaws on images taken at different timings, it is impossible to track flaw growth or changes.

  Further, in the prior art, processing or display suitable for image processing of the wheel side surface is not studied.

  In view of the above, the first object of the present invention is to provide a vehicle equipped with a reference mark detection method for specifying the position where the feature exists in relation to the absolute position in detecting a feature amount such as a flaw on a wheel side surface. It is to provide an inspection device.

  A second object of the present invention is to provide a vehicle inspection apparatus that can identify the same feature quantity from a plurality of wheel side surface images in different time relationships and manage the history of the feature quantity for which the position is specified. There is to do.

  A third object of the present invention is to provide a vehicle inspection apparatus provided with a process suitable for image processing of a wheel side surface or a display technique.

In the first aspect of the present invention, in order to achieve the first object of realizing the vehicle inspection device that detects the position of the feature amount,
Detect feature values on the wheels from the image of the side of the railcar wheel that was input, detect the coordinates of the wheel center from the image of the side of the railcar wheel that was input, and detect the reference mark from the image of the railroad wheel side that was input The coordinates are determined, and the position of the feature amount determined based on the coordinates of the wheel center and the coordinates of the reference mark is stored.

  The reference mark detection is performed by detecting a region having a high correlation value by template matching from an input image that holds a template for the reference mark.

  Further, the template for the reference mark is held as a plurality of templates rotated according to the rotation of the wheel, and an area having a high correlation value is detected by template matching from images input for each of the plurality of templates. Is detected.

  For reference mark matching, a plurality of reference mark candidate areas having a high correlation value are detected by matching detection using one type of template, and the reference mark candidate areas selected by a plurality of templates having different shooting conditions are matched, The highest position is the reference mark position.

  In the reference mark detection, character candidates are cut out from the input image, the intervals between the characters of the plurality of character strings existing on the wheel are stored as rules, and the character string candidates cut out using the rule of the arrangement interval of the character strings Among them, a character string candidate that matches the rule is detected, and the position of the reference mark is estimated.

  In the character candidate cutout, the mark position is estimated for each of the character string candidates cut out using a plurality of different threshold values, and set as the reference mark position.

  In the reference mark detection, a template for a reference mark is held, and a region having a high correlation value is detected from the input image by template matching to be used as a reference mark. Character candidates are cut out from the input image and exist on the wheel. The interval between characters of multiple character strings is stored as a rule, and the character string candidate that matches the rule is detected from the character string candidates using the rule of the character string arrangement interval, and the reference mark and the character string candidate mark The mark position is finally determined from the position candidates.

  In addition, when the correlation value of the reference mark position candidate is sufficiently high, the mark position determination has priority over the reference mark position candidate estimated by the mark position determination as the reference mark position.

  In addition, when there are multiple reference mark position candidates detected by reference mark matching and multiple reference mark position candidates estimated by mark position determination, the mark position determination gives priority to the position where the same position is detected or estimated. The reference mark position may be determined.

  Further, a plurality of different rules of the character arrangement interval rule are weighted, and based on the weighting, the priority is determined when determining the reference mark position from the plurality of reference mark position candidates when determining the mark position.

  In addition, when there are a plurality of reference mark position candidates detected by the reference mark matching and a plurality of reference mark position candidates estimated by the mark position determination, the mark position determination is an average of a plurality of estimated positions when there are a plurality of estimated positions in the vicinity. Alternatively, when a plurality of rules are weighted, the position calculated by taking the weighted average is determined as the mark position.

In the invention according to claim 12, in order to achieve the second object of realizing a vehicle inspection device that records the history of feature values,
Detect feature values on the wheels from the image of the side of the railcar wheel that was input, detect the coordinates of the wheel center from the image of the side of the railcar wheel that was input, and detect the reference mark from the image of the railroad wheel side that was input Stores the correspondence of the wheel identification information that determines the coordinates and identifies the shooting wheel when shooting the railway vehicle, and the position of the feature amount determined with reference to the coordinates of the wheel center and the coordinates of the reference mark as the wheel identification information If the feature value obtained from the image of the side surface of the railroad vehicle wheel that is stored in association with the input image is the same as the feature value that is considered identical from the feature value position information in the wheel information having the same wheel identification information Is additionally stored with identification information indicating the same feature amount.

  In addition, images of the same wheels taken on different dates are displayed side by side so that the same feature amount can be compared.

  Moreover, the relationship between the information regarding the traveling of the wheel and the information regarding the feature amount is displayed in comparison with the train operation information and the statistical information.

In the invention according to claim 16, in order to achieve the third object of realizing a vehicle inspection apparatus having a process suitable for image processing of a wheel side surface or a display method,
Detect feature values on wheels for multiple images of the input railcar wheel side, detect wheel center coordinates for the input railcar wheel side images, and mark reference marks for the input railcar wheel side images. The coordinates are determined by detection, and the corresponding positions of images taken at adjacent positions are obtained and connected.

  Further, the inter-image position collation calculates the wear amount of the wheel by detecting the center of the wheel on the image and the arc of the wheel, and calculates the corresponding point between the adjacent images.

  Further, the feature amount on the wheel is detected from the image obtained by polar coordinate conversion with the detected wheel center as a reference.

  ADVANTAGE OF THE INVENTION According to this invention, the method of detecting the reference mark for specifying the feature-value position of a wheel side surface by a relationship with an absolute position from a wheel side surface image can be provided.

  According to the present invention, it is possible to identify the same feature amount from a plurality of wheel side surface images in different time relations and manage their history.

  Furthermore, according to the present invention, it is possible to provide a processing or display technique suitable for image processing of the wheel side surface.

It is a block diagram which shows the structure of the 1st Example of the vehicle inspection apparatus by this invention. It is an example of the image which image | photographed the wheel surface at the inspection time 1. It is an example of the image which image | photographed the wheel surface at the inspection time 2. FIG. It is a figure which shows the example of the image input from the image input means. It is a figure which shows the example of the image extracted by the feature-value detection means. It is a figure for demonstrating the processing content of a wheel center detection means. It is a figure for demonstrating the processing content of a reference mark detection means. FIG. 5 is a diagram showing an example of data stored in a knitting / wheel correspondence storage means 5. It is a figure which shows the example of the information stored in a feature-value history storage means. It is a figure for demonstrating the flow of a process of a feature-value collation means. It is an example of the image input from the image input means. It is a figure which shows the example which added the record to the feature-value history storage means. It is a figure which shows the state of the feature-value history storage means when the process of a feature-value collation means is complete | finished. It is a figure which shows the example of the character string stamped on the wheel A. FIG. It is a figure which shows the example of the character string stamped on the wheel B. FIG. It is a block diagram which shows the structure of the 2nd Example of the vehicle inspection apparatus by this invention. It is the figure which showed the example of the image before polar coordinate conversion. It is the figure which showed the example of the image after polar coordinate conversion. It is a figure which shows the example of the character string candidate of an input image. It is a figure which shows the example of the character string candidate cut out by the character candidate cut-out means. It is a figure which shows the example of the character arrangement space | interval rule stored in a character arrangement space | interval rule storage means. It is a figure for demonstrating the flow of a process of a mark position estimation means. It is a figure which shows the comparison method of the position of a character arrangement interval rule, and the position of the cut-out character string. It is a figure for demonstrating the production method of a several template. It is a figure which shows the example of a manufacture information character string for demonstrating the effect of the Example which prepares several estimation rules. It is a figure which shows the example in which the reference mark estimated position of several rules corresponds. It is a figure which shows the example which one type of rule among multiple rules estimates incorrectly. It is a block diagram which shows the structure of the 3rd Example of the vehicle inspection apparatus by this invention. It is a figure which shows the example of the image image | photographed with the image imaging means. It is a figure which shows the example of the image divided | segmented before the process of the position collation means between images, and the example of the image which overlapped and connected the corresponding point after collation. It is a figure which shows the adjacent divided image after polar coordinate conversion. It is a figure which shows the adjacent divided image after polar coordinate conversion. It is a figure which shows the image which connected the divided image after polar coordinate conversion after collation. It is a block diagram which shows the structure of the 4th Example of the vehicle inspection apparatus by this invention. It is a figure which shows the example of the input screen for selecting the wheel displayed by the feature-value history display means. It is a figure which shows the example of a display of the history of the selected wheel. It is a block diagram which shows the structure of the 5th Example of the vehicle inspection apparatus by this invention.

  Examples of the present invention will be described below.

  Hereinafter, the vehicle inspection apparatus of the present invention will be described with reference to FIGS.

  The first embodiment describes the basic configuration and functions of the vehicle inspection apparatus according to the present invention, and will be described with reference to FIGS.

  First, FIG. 1 is a block diagram showing a schematic apparatus configuration of a first embodiment of the vehicle inspection apparatus of the present invention. The vehicle inspection apparatus of FIG. 1 includes an image input means 1, a feature quantity detection means 2, a wheel center detection means 3, a reference mark detection means 4, a composition / wheel correspondence storage means 5, and a feature quantity history storage means 6 And feature amount matching means 7.

  The components of the first embodiment will be described in detail in order. In the following description, for ease of understanding, the explanation is made using specific numerical values such as date, length, angle, etc., but these are all given as examples, and the numerical values themselves are exceptional. There is no meaning.

  First, FIG. 2 shows an example of an image obtained by photographing the wheel side surface to be processed in the present invention. FIG. 2 shows an example of images taken at different inspection times for the same wheel. FIG. 5A shows an image 21 taken at the inspection time 1, and FIG. 5B shows an image 22 taken at the subsequent inspection time 2. FIG. The wheel 23 and the wheel 23 'are the same wheel at different times. Marks indicated by 25 and 25 'are always present at one place in all the wheels.

  Here, a feature amount 26 exists on the wheel 23 photographed in the image 21. The position of the feature amount 26 can be defined by an angle 27 having 24 as the center of the wheel and the mark 25 as a reference. Here, the angle 27 is set to x °.

  A feature amount 28 and a feature amount 29 exist on the wheel 23 ′ captured after the image 22. Here, the positions of the wheels on the screen are usually different between the images 22 and 21 due to the rotation. Specifically, the position of the reference mark 25 of the image 21 on the screen 22 is moved to the mark 25 ′ on the screen 22.

  However, since the angle 2a formed by the reference mark 25 ′ and the center 24 corresponds to the existing angle x ° in the case of the image 21, the feature quantity identical to the feature quantity 26 on the screen 21 at this position is 22 can be identified as feature amount 28. If the angle 2b from the reference mark is y ° with respect to the feature amount 29, there is no feature amount on the image 21 at a position corresponding to this position. Therefore, the feature amount 29 will be continuously monitored as a new feature amount in the future.

  As described above, if a human is visually inspecting a wheel, which is a rotating inspection object, at an arbitrary timing, the position on the wheel can be collated. However, in order to perform this automatically, the wheel center detecting means 3 that detects the center of the wheel that dynamically changes on the screen and the mark position that is always in one place on all the wheels are detected and It is necessary to provide the reference mark detection means 4 as the reference position of the angle, identify the feature amount position generated on the wheel, and compare the change in size and shape of the same feature amount. It should be noted that a mark assumed as a reference is a unique mark that always exists at one place for each wheel, such as a mark of a wheel manufacturer.

  Based on the above assumptions, the components of the first embodiment will be described in order. The image input unit 1 in FIG. 1 is a unit that captures and inputs an image to be inspected. Input images as shown in 21 and 22 of FIG.

  The feature amount detection unit 2 is a unit that detects a feature amount from the image captured by the image input unit 1. As for this realization means, for example, if the feature amount is photographed with a lower luminance than the wheel surface (black), the feature value is obtained by binarization with a threshold value set between the average luminance of the wheel surface and the average luminance of the feature amount. A method such as extracting the quantity region black is used.

  FIG. 3 is a diagram for explaining the processing contents of the feature amount detection means 2. FIG. 2A shows an example of the image 31 input from the image input means 1, and there are feature amounts 32 and 33 on the wheel 32. This is extracted by the feature amount detection means 2, and the start point and end point coordinates on the screen are obtained for each feature amount with reference to the lower left point 39 on the screen, for example, as indicated by 35 in FIG. . For example, for the feature amount 37, the coordinates of the start point 3a and the end point 3b are obtained, and for the feature amount 38, the coordinates of the start point 3c and the end point 3d are obtained, and the coordinates that can be managed on the screen are obtained. Here, the coordinates described in FIG. 3 are only coordinates on this image, and needless to say, they are not the absolute coordinate positions obtained from the reference marks and wheel centers described in FIG.

  In the image range acquired in FIG. 3, the axle region indicated by 3e in the wheel that does not need to detect the characteristic amount is also shown. The processing unnecessary area is subjected to mask processing as an area that does not need to be processed in advance and is not subjected to feature amount detection processing, thereby preventing erroneous detection of unnecessary areas. The method of obtaining the range of the axle region can be obtained from the center of the wheel in the image obtained by the wheel center detecting means 3 described later and the radius of the known axle region.

  The wheel center detecting means 3 is means for obtaining the center coordinates of the wheel on the image in order to uniquely obtain the position of the detected feature quantity on the wheel. For example, as disclosed in Japanese Patent Application Laid-Open No. 6-167313, this means can be obtained by means such as detecting the edge of the arc of the wheel by image processing and obtaining the center position from approximation with the arc of the circle. it can. FIG. 4 is a diagram for explaining the processing content of the wheel center detecting means 3, and as shown in this figure, the obtained center position 43 is the same as the position of the feature quantity detected by the feature quantity detecting means 2. This is grasped by the relative coordinates from the start point 44 in the image, and is stored in the memory.

  The reference mark detection means 4 is a means for detecting a specific mark present on the wheel as a reference position in the circumferential direction of the wheel. One of the embodiments for realizing the reference mark detection means 4 is a method for obtaining a position where the texture is most similar to the template by using, for example, pattern matching.

  FIG. 5 is a diagram for explaining the processing contents of the reference mark detection means 4. If the specific mark is a texture mark such as 51 in FIG. This is a method of detecting a position having a large correlation value as a mark position by matching with information on an image held as a template. In the image of 52, the position having a high degree of coincidence with 51 is the position centered on 53 in the figure, so that the relative coordinates based on the lower left reference point 54 in the image are the same as the wheel center position and feature position. To grasp the position of the reference mark and store it in the memory.

  In the case of a disk-like object such as a wheel, the mark is often attached along an arc centered on the center of the wheel, and the direction of the mark in the image can be rotated 360 °. It is thought that there is sex. Therefore, as a means of pattern matching, for template 51, a plurality of templates that are rotated little by little within a 360 ° range are created, and all templates are matched to adopt the position of the maximum correlation value. Can be used.

  Next, the knitting / wheel correspondence storage means 5 will be described. The knitting / wheel correspondence storing means 5 is a means for storing the correspondence between the ID for identifying the knitting and the ID of the wheel incorporated in the knitting. An example of data stored in the knitting / wheel correspondence storage means 5 is shown in FIG. In FIG. 6, the column direction indicates storage items, and the row direction is an example of a data record.

  In FIG. 6, a column 61 is an organization ID that can identify a series of organizations, and is an ID that can uniquely identify an organization. A column 62 is a code for identifying a side surface (left / right) with a wheel, where L is left and R is right. A column 63 is an array element number, and indicates an array element number counted from the top of the side surface specified in the column 62. A column 64 is a wheel ID, which is a number that can uniquely identify a wheel.

  Also, because the wheel may be removed from the train and replaced, the date the wheel was attached to the train on the “start date” shown in 6b and the wheel removed from the train on the “end date” in 6c. The stored date is also stored. It is also possible to refer to the history of the wheels that have been attached to the position in this way. The correspondence between the latest knitting and the wheel can be acquired by referring to a record in which the “end date” is blank.

  For example, various data are stored in the knitting / wheel correspondence storage means 5 under the above-mentioned promise, and data are individually stored in the rows 65 to 70. Among these, for example, the data record shown in the row 65 indicates that the frontmost wheel on the left side of the knitting with the knitting ID “1” has the wheel ID “AAA”. Similarly, the data record shown in the row 70 indicates that the third wheel from the top on the right side of the composition of the composition ID “1” is the ID “BBD”. As described above, the train / wheel correspondence storage means 5 can acquire the train wheel ID using the train train ID, the left and right side surfaces, and the array number as keys.

  The feature quantity history storage means 6 is means for storing information relating to the history of the feature quantity generated on the wheel. FIG. 7 shows an example of information stored in the feature amount history storage means 6. In the figure, 71 to 78 in the first row are examples of storage items, and the rows 7a and 7b are examples of data.

  In FIG. 7, a column 71 is a date when the feature amount is detected. A column 72 is a wheel ID in which a feature amount exists. A column 73 is a feature amount ID for specifying the feature amount. A column 74 is an angle of the feature amount from the reference mark for specifying the position of the feature amount. A column 75 is a position from the center for specifying the position of the feature amount. Column 76 is the length of the feature quantity. A column 77 is the starting point coordinates of the feature amount on the image. Column 78 is the end point coordinates of the feature amount on the image. A column 79 is a name of an image in which the feature amount is captured.

  In the feature quantity history storage means 6, various data are stored under the promise as described above. For example, in the data of the row 7a, the data was taken on "December 10, 2008" (column 71). In the image, on wheel ID `` AAA '' (row 72), the feature amount is 30 mm (row 76) at a position where the angle from the reference is 33 ° (row 74), the position from the center is 256 mm (row 75) It shows that there is. From the column 73, this feature amount is managed by a unique ID “AAA-1”. Further, from column 77 to column 79, information indicating that the feature amount is on the start point coordinates (520, 200) and the end point coordinates (523, 215) on the image name “img090210AAA” is shown.

  As described above, the information stored in the feature quantity history storage means 6 is the storage means (not shown) for managing the date and position / size when the feature quantity of the wheel is detected, and for managing the photographed image. If there is, there is a means for specifying the position where the feature amount is reflected.

  The feature quantity matching means 7 identifies the feature quantity detected by the feature quantity detection means 2 using the positional information and the feature quantity detected in the past examination, and if there is a feature quantity that is considered to be the same, it is the same It is a means to give ID.

  A processing flow of the feature amount matching unit 7 will be described with reference to FIG. In this process, information in the knitting / wheel correspondence storing means 5 and information in the feature amount history storing means 6 are used.

  In this flowchart, in a process 81, a wheel ID is acquired from the knitting / wheel correspondence storing means 5. In this process, when the wheel image is input, the composition ID is input in advance. Alternatively, when the organization enters the image acquisition position of the inspection site, etc., the schedule schedule data is stored in advance, and when the image input is started, the ID of the organization that enters the inspection site at that time, and The wheel ID is obtained for each photographed image from the wheel array. Here, the wheel ID is x.

  In the process 82, the feature quantity counter I is cleared to zero. In process 83, an angle from the reference of the I-th feature amount is calculated. In process 84, the feature amount counter I is incremented by one. In process 85, it is determined whether or not all feature quantities have been processed. If all feature quantities have been processed, the process returns to process 86. If there is an unprocessed feature quantity, the process returns to process 82, and all feature quantities have been processed. The process is repeated until the angle calculation process is completed.

  During the above-described series of cyclic processes, the process 83 calculates the angle of the feature quantity from the reference mark, and stores the result in the feature quantity history storage means 6. FIG. 9 shows an example of an image obtained from the image input means, and the processing contents of processing 83 will be described with reference to FIG. In the example of the image of the wheel 91 in FIG. 9, by the process 83, the positions 92 and 93 of the feature amount detected by the feature amount detection means 2, the wheel center 94 detected by the wheel center detection means 3, and the reference mark detection means 4 The position 95 of the reference mark detected by the above is derived, and the angle is calculated.

  In the process 83, when obtaining the angle from the positional relationship on the image, it is assumed that the direction of the camera is perpendicular to the wheel surface. However, in many cases, the camera and the wheel surface are not perpendicular but inclined. it is conceivable that. In that case, from the angle information of the camera and the wheel, coordinate conversion is performed by camera calibration, the camera orientation is perpendicular to the wheel surface, converted into a confronted image, and on the confronted image, It is effective to obtain the angle between the reference mark and the feature amount.

  In the process 86, information on the feature quantity detected in the previous inspection of the wheel x is extracted from the feature quantity history storage means 6. In process 87, the counter J that counts the feature amount detected in the previous inspection of the wheel x is cleared to zero. In the process 88, the feature quantity I that matches the feature quantity j detected in the previous inspection is searched from those detected this time. In the process 89, the feature quantity ID of the feature quantity j is set to the feature quantity I that matches the feature quantity j. In process 8a, it is determined whether or not all the feature values detected in the previous inspection have been processed. If all the feature values have been processed, the process 8b is performed. If there is an unprocessed feature value, the process 88 is performed. Return. In process 8b, a new feature quantity ID is assigned to a new feature quantity that has not been given a feature quantity ID in process 88 to process 8a among the feature quantities detected this time.

  In the process 88, whether or not the feature amounts match is determined by comparing the angle from the reference mark of the column 74 stored in the feature amount history storage means 6 and the position from the center of the column 75. A match or the closest value is determined as a match.

  The above-described series of processing flow in FIG. 8 will be described using specific data. FIG. 9 is an example of the above-described image input from the image input means 1, and this image is the first image taken on the left side of the composition having the composition ID “1”. In the figure, 91 is a wheel, 92 and 93 are detected feature quantities, 94 is the center of the wheel, and 95 is a detected reference mark.

  In process 81, the wheel ID is obtained from the composition / wheel correspondence storage means 5. From the example of the knitting / wheel correspondence storage means 5 in FIG. 6, the knitting ID is “1”, the first wheel arranged on the left side and the arrangement is the data in row 65, so the wheel ID is “AAA”. get.

  In process 83, the angle of the 0th feature amount is calculated. At this time, it is necessary to determine a point representing the feature quantity, but the center of gravity obtained from the start and end points of the feature quantity is used. The angle between the calculated feature quantity 92 and the reference mark is 33 °, the position from the center is 254 mm, and the coordinates of the start and end points are added to the feature quantity history storage means 6. Fig. 7 records the feature value of the wheel ID "AAA" detected on December 10, 2008. The record with the results of the vehicle inspection conducted on February 10, 2009 is added in Fig. 10 Shown in line 101. The date is registered from the time when the information is acquired, and the wheel ID and the input image name are registered.

  In the example of FIG. 9, since the number of feature quantities is “2”, the process returns to process 83 in the determination of process 85. Similarly, the angle between the feature amount 93 in FIG. 9 and the distance from the center are calculated, and the coordinates of the start and end points are added to the feature amount history storage means 6. 102 records were added. At this time, the feature amount ID has not been given yet.

  Since the processing of all the feature values of the wheel “AAA” has been completed, the processing shifts to processing 86. The feature quantity of the previous inspection of the wheel “AAA” is extracted from the feature quantity history storage means 6. From FIG. 10, when the record of the feature amount of the previous inspection of the wheel “AAA” is extracted, it becomes one in row 7a.

  In process 88, a matching feature quantity is searched for the feature quantity “AAA-1” in the row 7a, which is the feature quantity of the wheel “AAA” at the time of the previous inspection. The feature amount detected from the image this time is registered in the rows 101 and 102 of the feature amount history storage means 6, but the record with the angle of 66 ° coincides with the record in the row 101. The feature amount ID is given as “AAA-1”, assuming that the feature amount matches the record.

  Since the record of the feature quantity at the time of the previous inspection is this one, the feature quantity corresponding to the row 102 in FIG. 10 is not detected the previous time but is newly detected this time, and the new feature quantity ID “ AAA-2 ”is granted.

  FIG. 11 shows the state of the feature amount history storage unit 6 when the processing of the feature amount matching unit 7 is completed. The difference from FIG. 10 is that a feature ID is newly assigned. As a result, by obtaining the invariant position of the wheel from the position of the mark obtained as the reference position and the center position, the feature amount detected when the reference position in the image is different at the time of different inspections. You can collate.

  The above is the first embodiment of the vehicle inspection apparatus. According to this embodiment, even if the wheel position in the image is changed by the wheel center detecting means 3 and the reference mark detecting means 4, the position on the wheel is determined by the position from the reference center and the angle from the reference mark. Since the coordinate system can always be uniquely determined, it is possible to manage the feature history without being conscious of the direction of the wheel when taking an image.

  Next, a second embodiment of the vehicle inspection apparatus will be described. In the second embodiment, various devices for accurately detecting the reference mark position necessary for determining the absolute position of the feature amount will be introduced with reference to FIGS.

  The second embodiment of the vehicle inspection apparatus is devised in that the position of the reference mark is estimated using a rule for the arrangement interval of the character strings including the reference mark, in addition to the mark detection by pattern matching. As an explanation of the present embodiment, a process for estimating the position of a reference mark by detecting the arrangement of character strings of manufacturing information stamped on the wheels of a railway vehicle will be described as an example.

  As a premise of the second embodiment, it is assumed that an example of a character string of manufacturing information to be assumed is as shown in FIG. The premise of the present embodiment is that there is one reference unique mark at a position on the circumference and that the number of digits in the character string is constant. 12A shows an example of a character string 121 stamped on the wheel A, and FIG. 12B shows an example of a character string 124 stamped on the wheel B. Since these are manufacturing information about the wheels, the contents of the characters differ depending on the wheels, but it is assumed that one or more places always include unique marks or characters. In the illustrated example, the marks 122 and 125 in the character string 121 and the character string 124 have the same shape and are stamped on any wheel, for example, a manufacturer's mark. On the other hand, the character strings indicated by 123 and 126 are often variable depending on the wheels, such as the manufacturing date and the manufacturing number.

  In the vehicle inspection of the present invention, since inspection at the time of maintenance is assumed, noise such as dirt adheres to the wheels to be inspected, and the lighting environment at the time of image acquisition is difficult to control appropriately. This is a situation where it is difficult to obtain all the character strings of the manufacturing information on the image. Therefore, a method for detecting the position of a unique mark in the character string of the manufacturing information even if the content of the character string of the manufacturing information cannot be recognized will be described in this embodiment.

  In the present embodiment, description will be given by taking as an example the estimation of the positions of the marks 122 and 125 in FIG. FIG. 13 is a block diagram illustrating the configuration of the second embodiment. As the configuration of the reference mark detection unit 4 in the configuration of the first embodiment in FIG. 1, the character candidate cutout unit 8, the character arrangement interval rule storage unit 16, the mark position estimation unit 9, the mark position determination unit 10, and the reference The mark matching means 17 is added.

  In the following description of the second embodiment, the newly added configuration will be described in order. First, the character candidate cutout unit 8 is a unit that cuts out a character or a specific pattern from the image input from the image input unit 1. Hereinafter, although described as “character candidates”, several types of identifiable patterns may be used.

  At this time, it is possible to process the input image as it is. However, in order to facilitate coordinate management during image processing, the input circular wheel image is converted into a developed image by polar coordinate conversion and then processed. It is effective. . This conversion process converts the input image position coordinates (x, y) to (r, θ) (r is the radius of the circle and θ is the angle of the circumference) with reference to the center of the circle in the image. Then, it can be executed by converting θ into an orthogonal coordinate system of x-axis and r as y-axis.

  FIG. 14 shows examples of images before and after polar coordinate conversion. In FIG. 14A, 141 is an image before polar coordinate conversion, and 146 in FIG. 14B is an image after polar coordinate conversion. With the wheel 142 in the image 141, the length in the radial direction connecting the outer peripheral direction of the wheel from the wheel center 143 detected by the wheel center detecting means 3 is set to 144. This corresponds to the length r in the circular system direction in the coordinate system before conversion. Reference numeral 145 denotes an angle centered at 143, which corresponds to the circular angle θ in the coordinate system before conversion.

  The converted image 146 is obtained by converting into a coordinate system with the center 143 of the wheel as the center and the radial length r represented by 144 as the x-axis and the angle 145 as the y-axis. Since image processing is usually handled in the xy plane coordinate system in many cases, there is an advantage that the coordinate data of the image processing result can be easily handled by converting the polar coordinate before the image processing into the xy coordinate system. The processing of the second embodiment is performed on the assumption that polar coordinate conversion has been performed, but it goes without saying that the same processing can be performed when conversion is not performed by replacing the x coordinate in the description with an angle and the y coordinate with a radial length.

  Returning to the description of the processing of the character candidate cutout means 8. FIG. 15 shows an example of character string candidates cut out by the character candidate cutout means 8. In FIG. 15A, 151 is an input image, and 15 in FIG. 15B is a character string candidate cut out by the character string cutting means. As a character string cut-out method, for example, if the character is lower in brightness than the wheel color (black), a value between the average value of the wheel background color and the average value of the character color is appropriately determined as a threshold, By binarizing, it can be realized by processing such as extracting a character string area from the background area. In addition, since binarization alone is used, characters are interrupted and the distinction from noise is unclear. Therefore, noise such as isolated points is removed by performing expansion / contraction processing, and the outline of the character area is defined. Extract clearly.

  Next, the character arrangement interval rule storage means 16 will be described. FIG. 16 shows an example of the character arrangement interval rule stored in the character arrangement interval rule storage means 16. This embodiment is based on the premise that character strings such as manufacturing information are equally spaced.

  Reference numeral 161 in FIG. 16 shows an example of a character arrangement rule for manufacturing information in this embodiment. As described with reference to FIG. 12, 16a in FIG. 16 is a reference mark, and it is assumed that there is always one mark on all wheels. On the other hand, the rear character string 16b is a sequence of numbers or symbols different for each wheel.

  As an example of the character arrangement rule, a rule (16d) for character arrangement of 8 characters on the left and right with reference to the character 16c in FIG. 16 is shown. Using the center of the character 16c as a reference, the interval 162 between the right adjacent character, the interval 163 with the left adjacent character, the interval 164..., And the interval with all eight digits up to 168 are defined. The interval is defined by the frequency before polar coordinate conversion or the number of pixels after polar coordinate conversion. Further, an interval 169 between the reference character 16c and the reference mark 16a is also defined. If the positional relationship between the digit arrangements as rules can be defined, the reference character selection method and the like are not necessarily based on this example.

  In addition, the character sequence selected as a rule may be anything unique as long as it is unique among all manufacturing information character strings, and even the entire manufacturing information character string shown in 16b is a partial sequence as shown in 16d. But it doesn't matter. It is assumed that the character spacing is a sequence that can uniquely identify the position in the manufacturing character string. As an inappropriate example, for example, the 4-digit arrangement shown in 16e partially matches the 8-digit arrangement of 16d, so such a character arrangement is avoided.

  One or more types of character arrangement interval rules defined as described above are stored in the character arrangement interval rule storage means 16.

  Next, the process of the mark position estimating means 9 will be described. The mark position estimating unit 9 uses the character string arrangement interval rule stored in the character arrangement interval rule storage unit 16, and the character string candidate that matches the rule from the character string candidates extracted by the character string candidate extraction unit 8. And the position of the reference mark is estimated from the interval between the rule character string and the reference mark. Among the character string candidates cut out by the character string candidate cutout means 8, a character line whose interval matches the character intervals 162 to 168 in FIG. 16 is detected, and the position of the reference mark is determined by the interval 169 between the character line and the reference mark. The mark position estimating means 9 is estimated.

  The process flow of the mark position estimating means 9 will be described with reference to FIG.

  In process 171, the counter I that counts the number of types of rules is cleared to zero. In process 172, the counter J that counts the number of characters cut out by the character candidate cutout means 8 is cleared to zero. In the process 173, the character arrangement based on the jth character candidate extracted by the character candidate extraction means 8 is compared with the arrangement of the rule I, and the position error in the x direction is calculated. In process 174, the counter j of the number of characters cut out by the character candidate cutout means 8 is incremented by one. In process 175, it is determined whether or not the counter j of the number of characters cut out by the character candidate cutout means 8 exceeds the number of character string candidates. If it exceeds, the comparison with rule I based on all character candidates is completed, and the process proceeds to process 176. If not, the process returns to process 173, and the jth character string is processed.

  In process 176, a process of acquiring the position of the character candidate with the smallest error is performed for rule I, and in process 173, the value of j that minimizes the error is obtained from the case where the jth character candidate is used as a reference. . In process 177, the position of the reference mark is estimated as a relative position from the position of the character candidate that minimizes the error obtained in process 176. In process 178, the counter i for counting the number of types of rules is incremented by 1, and the process proceeds to the next rule. In process 179, it is determined whether or not the counter I that counts the number of types of rules exceeds the number of rule types. If it exceeds, it is assumed that the processing for all the rules is finished, and the processing is finished. If not exceeded, the process returns to the process 172 to process the next rule.

  The above is the processing flow of the mark position estimating means 9. The content of the processing will be described using a specific example. FIG. 18 is a diagram illustrating a method of comparing the character and interval rule and the position of the extracted character string. In the figure, 181 is an example of a character arrangement rule, 182 is a reference character in the rule, and 183 is a reference mark position estimated in a relative position to the reference character 182 in this rule.

  On the other hand, the character candidates cut out by the character string candidate cut-out means 8 are shown at 152. In the figure, for convenience of explanation, the sequence numbers (0) to (14) are displayed at the bottom of the character candidates.The character candidates (0) to (14) are used as reference positions, and The contents of the process 173 and process 174 are compared in order with the arrangement of the character arrangement rules shown.

  In the example of FIG. 18, the rule reference character 182 and the center x coordinate of the character candidate 184 in the array (5) are matched, and the x coordinate of the other 7-digit character is compared with the closest character candidate. Then, the absolute value of the error is accumulated for 7 digits, and the error based on the jth character candidate of rule I is calculated. The processing of 173 is calculated while the reference position is advanced from 0 to 14 by the counter of j.

  Since the character arrangement rule of 181 is an 8-digit uniform character arrangement, when error evaluation is performed by matching the reference character 182 of the rule in order from the character candidates (0) to (14), the arrangement shown in FIG. (5) When combined with 184, the error is the smallest. Therefore, the position of j = 5 is set to a position that matches the rule of 181 and the position of the reference mark is estimated based on this position.

  In FIG. 18, the interval indicated by 169 is the interval between the reference character 182 and the reference mark 183 of the rule 181, and therefore when the character candidates cut out are overlapped, the estimated position becomes the position 185 in the drawing. . The position 185 is the estimated position of the reference mark according to the rule 181.

  Similarly, when there are a plurality of rules in the character arrangement interval rule storage unit 16, the above process is repeated for the number of rules, and an estimated position that minimizes the error is obtained for each rule.

  In the above embodiment, the estimated position is the position with the smallest error, but in some cases, the extracted character candidate is missing due to the influence of shadow, noise, blurred character, etc. It may not be cut out as a character string as shown. In such a case, there is a possibility of detecting different character arrangements even with the smallest error, so an error threshold is set, and the error is below the threshold, and the position of the minimum value is adopted as having a high certainty of estimation. Examples can also be taken.

  The reference mark matching unit 17 is a process for obtaining the position of the reference mark having a high correlation value by pattern matching, similarly to the reference mark detection unit 4 of the first embodiment. In the first embodiment, in order to perform template matching on the input image, a plurality of reference mark templates having different orientations depending on the circumferential position are prepared and matched. As in 146 of b), the character string on the circumference can be converted into a character string image arranged in the horizontal direction. When pattern matching is performed on the wheel image after the polar coordinate conversion, the direction of the reference mark is constant no matter where on the image, and therefore one type of matching template can be implemented.

  On the other hand, when the reference mark is an inscription stamped on the wheel, the illumination angle changes when the position of the inscription changes, so the direction in which the shadow of the inscription changes. Even if the stamp is the same, the correlation value of matching also decreases if the direction of the shadow of the template is different. Therefore, in order to obtain a high correlation value by matching regardless of the position of the reference mark, multiple marks at various positions on the screen are used. An embodiment in which an image is used as a template and matching is also possible.

  FIG. 19 shows a method for creating a plurality of templates. In FIG. 19, 311, 313, and 315 are images after the polar coordinate conversion of the wheel, and the reference marks exist at different positions in the x direction in the image. These reference mark areas 312, 314, and 316 are cut out and created in advance as a plurality of templates as indicated by 317.

  Matching is performed using the plurality of templates created in this way, and the position where the correlation value is maximized is set as a reference mark candidate position by matching. At this time, the entire detection area may be searched for all templates, but the processing load increases as the number of templates increases.For example, the entire detection area is searched using the template adopted from the vicinity of the center in the image. Alternatively, a method of selecting a plurality of candidate positions having a high matching correlation value and calculating a correlation value by matching each of the selected candidate positions with a plurality of templates may be employed.

  Next, the mark position determining means 10 having the configuration of the second embodiment will be described. The mark position determining means is a means for finally determining the mark position from the plurality of mark position candidates obtained by the reference mark detecting means 4 and the mark position estimating means 9.

  In this means, the most probable reference mark position from the detection position of the reference mark detection means 4, the correlation value of the matching, the estimated position by a plurality of rules obtained by the mark position estimation means 9, the estimation error, etc. It is a process to determine.

As a determination condition for determining a mark position from a plurality of mark position candidates, for example, there are the following methods.
Since the correlation value by matching indicates the degree of similarity of the texture of the mark, a position where the correlation value by matching is high has a high certainty that it is a reference mark, and the correlation value of the detection position by the reference mark detection means 4 is greater than or equal to a threshold value. In some cases, the detection position is unconditionally adopted as the mark position. If a plurality of mark candidate positions are the same position, that position is adopted as the mark position.
When there are a plurality of estimation rules, weighting is performed between the estimation rules by priority. As a method of assigning priorities, character arrangement rule patterns are characteristic, and those with few false estimates increase the weight, while character arrangement rules that are easily misdetected due to noise or the like have a low weight. And so on. If the estimated positions of the marks differ among the plurality of rules, the most probable reference mark position is determined in consideration of these weights and errors.

  As described above, if the mark candidate position by matching and the estimated mark candidate position are obtained and the mark position is finally determined, the candidate position by matching is adopted when the mark is clearly visible on the image. can do. Also, if the mark is in the shade of lighting or is faint and faint, it does not become obvious on the image, so the estimated position is estimated by estimating from the sequence of character strings other than the mark. Can be adopted.

  Further, as an effect of the embodiment in which a plurality of estimation rules are prepared in the character arrangement interval rule storage means 16, as indicated by 202 in the character string 201 of FIG. 20, the character string of the manufacturing information is partially thin. If it cannot be seen due to shadow hiding or the like, the rule shown in 203 fails to estimate by the rule because part of the 8-digit number cannot be extracted as a character candidate.

  If the only prepared rule is 203, the reference mark detection by estimation will also fail, but the rule shown in 204 is also prepared, and if the estimation by this is successful, the reference mark position can be specified. . As described above, by preparing a plurality of rules, even when a part of the manufacturing information character string is hidden, the success probability of specifying the reference mark position by estimation can be increased.

    Further, as an effect of preparing a plurality of rules in the character arrangement interval rule storage means 16, there is an effect of adopting a mark candidate position where a plurality of estimated positions match. As shown in FIG. 21, the character candidate cut out by the character candidate cutout unit 8 from the character string of the manufacturing information shown in 211 is assumed to be 212.

  In contrast, the character arrangement rules are assumed to be three types 213, 214, and 215 in the figure. In each of the rules 213, the error is minimized at the position with the character candidate 216 as a reference, and the reference mark estimated position is 219. The rule 214 has a minimum error at a position based on the character candidate 217, and the reference mark estimated position is 21a. According to the rule 215, the error is minimized at the position based on the character candidate 218, and the reference mark estimated position is 21b. In this example, since the character candidates can be cut out correctly, each rule matches the appropriate position, and as a result, the plurality of estimated positions 219, 21a, and 21b match.

  This is an example in which the estimation is successful, but FIG. 22 is an example in which noise such as a flaw exists in addition to the manufacturing information and an erroneous estimation occurs. In FIG. 22, 221 is a character string of manufacturing information, and 222 is manufacturing information, whereas 223 is a flaw. The character string candidate cutout means 8 cuts this out 224, but a scratch other than a true character is cut out as in 22e if the character and color are similar.

  For the 224 character candidates, the position with the smallest error is obtained with the three types of rules 225, 226, and 227 as in FIG. 21, and the rule 225 and rule 226 match the correct position, and the reference mark estimated position is 22b and 22c and the matching positions are estimated. On the other hand, the rule 227 is originally correct to match the character candidate 22f, but the cut-out character candidate 22g is similar to the 4-digit character array 22e, so that the character candidate list based on 22a is better. There are cases where the error is reduced.

  In this case, the estimated position of the reference mark is 22d, which does not match the estimated positions of the other two rules. However, when the plurality of estimated positions are the same position based on the above-described reference mark position determination conditions, the positions are adopted as the mark positions. In this case, the mark positions are determined by the positions 22b and 22c. Priority will be given.

  In the above description, the multiple estimated positions are assumed to be “when they are the same position”. However, there are cases where errors occur in the character width, center position, etc. Even if they match correctly, an error may occur in the estimated position.

  For this reason, when determining as the same position, a range in which an error is taken into consideration is provided. For example, even when an error occurs in the estimated position for one character, an operation such as determining the same position is also conceivable. In that case, there is an operation such as taking an average value instead of adopting any of a plurality of estimated positions. In addition, when priority is weighted for each rule, an operation such as taking a weighted average considering the weight may be considered.

  In addition, when character candidates are cut out by binarization by the character candidate cutout unit 8, binarization is performed using a plurality of threshold values to cut out character candidates, and the mark position estimation unit 9 performs estimation from cutout results based on the respective threshold values, Reference mark estimated position candidates can also be obtained.

  In this case, there is a possibility that the mark position estimating means 9 may obtain the reference mark position candidates of “number of rule types × threshold number” types. From these multiple candidate positions, the mark position determining means 10 uses one reference position. Determine the mark position. In this way, by cutting out character string candidates by changing the threshold, even when the character string is thin on the image, it can be cut out as a character string candidate by binarization with a low threshold, or when there is a lot of noise, There is an effect of not cutting out noise at a high threshold. On the other hand, the possibility of cutting out a lot of noise also increases, but as shown in FIG. 22, by giving priority to the estimation results based on a plurality of rules, erroneous estimation due to a single noise can be eliminated.

  As described above, in the case where there is a concern about erroneous estimation due to noise, using as many rules as possible has an effect of preventing erroneous estimation in which the estimation result of the wrong position is adopted as the final mark position.

  As described above, in the second embodiment, even when the reference mark detection unit cannot detect the shape of the reference mark itself due to poor illumination conditions or noise such as dirt, it is estimated by the character arrangement interval rule of the manufacturing information. Thus, it is possible to estimate the reference mark position and collate the feature amount.

  Next, a third embodiment of the vehicle inspection apparatus will be described. The third embodiment provides a process or display method suitable for image processing of a wheel side surface, and will be described in detail with reference to FIGS. In this embodiment, the side surface of the traveling wheel is sequentially divided and photographed by a plurality of photographing devices, the feature amount is detected from each divided image of the wheel, and one reference mark position is detected from the divided image. And it is an Example which specifies a feature-value position about the perimeter of a wheel side surface.

  FIG. 23 is a block diagram showing a third embodiment of the vehicle inspection apparatus. Compared with the second embodiment, a new configuration is an image photographing means 11 and an inter-image position matching means 12.

  Mainly, the new configuration will be described. The image photographing means 11 is a means that is installed beside the track indicated by 232 and photographs the side surface of the traveling wheel 231. The image photographing means 11 includes a plurality of cameras 233 to 236. Image capturing by the image capturing unit 11 is executed based on a wheel detection signal from a wheel detection unit (not shown) installed toward the track side for each of the cameras 233 to 236, for example.

  An example of a photographed image is shown in FIG. Since the upper part of the wheel mounted on the railway vehicle is hidden behind the carriage part and cannot be seen, a part of the circumference is photographed. Further, the position of the vehicle detection means is adjusted so that the image of the wheel in the image can be taken at a position in contact with the track at the center of the screen.

  The cameras 233 to 236 constituting the image photographing means 11 are installed at an interval in consideration of the length of the arc of the wheel so that the entire circumference of the rotating wheel can be photographed by any camera. For example, in the case of dividing and photographing with four cameras as in this embodiment, the cameras are installed at intervals of the arc length corresponding to a quarter of the wheel.

  The image input means 1, the feature amount detection means 2, the wheel center detection means 3, the reference mark detection means 4, the character candidate cutout means 8, and the mark position estimation means 9 of the configuration of the third embodiment are the first implementation. This is carried out in the same manner as described in the example and the second embodiment. The difference between the first embodiment and the second embodiment is that the marks are not always present in the image because the wheels are divided and photographed. For this reason, the reference mark detection unit 4 and the mark position estimation unit 9 perform processing for detecting and estimating a position that is most likely to be a reference mark in each image.

  The inter-image position matching means 12 is a means for obtaining the corresponding positions of adjacent images among a plurality of images taken by dividing the same wheel and connecting the images. The purpose of this means is that since the image is divided, the mark position estimating means 9 and the feature amount matching means 7 can appropriately determine the positional relationship when the two points for calculating the positional relationship are separated into different images. This is to allow calculation.

  If the wheel diameter is constant, this means is always processing with a fixed position as a corresponding point. For example, when installing multiple cameras at an arc distance equivalent to 90 ° rotation of a wheel with a diameter of L, the distance between adjacent image center points is exactly 90 °. Therefore, the corresponding points may be obtained so that the angle between the center points of the image is 90 °. If the wheel diameter is shortened due to wheel wear or the like, if the wheel diameter after wear is α, the angle θ between the center points of the image can be obtained by the following calculation formula (1).

θ = 90 ° x (L / α) (1)
When the wheel diameter changes due to wear or the like, the arc of the wheel in the image is obtained by edge detection or the like, the distance between the center of the wheel and the arc is obtained on the image, the length α of the diameter is obtained, Find the interval between the central points using Equation 1.

  With the above method, corresponding points between images are obtained. FIG. 25 shows an example of a wheel image that is divided and photographed before processing by the inter-image position matching means 12, and an example of a wheel image that is connected after overlapping corresponding points. In the figure, 251, 253, 255, and 257 are wheel images taken separately by the cameras 233 to 236 in FIG. 23, and 252, 254, 256, and 258 are the centers of the screens 251, 252, 253, and 254, respectively. Is a point. 25d is an example of a wheel image in which positions between images are collated and connected, but an interval 259 between the center points 252 and 254 of adjacent images, an interval 25a between 254 and 256, and 256 and 258 The intervals 25b, 258, and 252 are set to the angles θ calculated by the equation (1).

  Similarly, the divided images after the polar coordinate conversion can be collated based on the above concept. FIG. 26 is a diagram illustrating a process of connecting the divided images after the polar coordinate conversion after collation. FIGS. 9A and 9B show adjacent divided images 261 and 262 after polar coordinate conversion, and FIG. 10C shows a connected image 265 after collation. As is clear from this figure, as in the case before the conversion, the positions of the images are collated so that the interval 266 between the center 263 of the screen and the distance 264 is the angle θ. In the example of FIG. 7C, 267 is an overlapping portion between images. For example, if the position of the x coordinate of 268 on the image is '1000' on the image 261 and '0' on the image 262, the position of the x coordinate 'xx' on the image 261 is Then, it is calculated as x coordinate 'xx-1000'.

  If the mark position estimated by the mark position estimation means 9 is not included in the same image, the estimated position in the adjacent image is obtained by converting the position in the adjacent image.

  The above is the description of the processing of the inter-image position matching unit 12.

  The mark position determining means 10 in the third embodiment is different from the second embodiment in that a mark position candidate detected by the reference mark detection means 4 in each of a plurality of divided images and a mark position estimation means 9 This is a point for determining one mark position from a plurality of estimated mark position candidates. Although there is a possibility that the number of mark position candidates increases, the determination method is the same as in the second embodiment, and one place is determined.

  Similarly, the feature amount matching means 7 also checks the position of the feature amount based on the positional relationship between the mark position and the feature amount in the image after the position between the images is verified.

  The above is the third embodiment of the vehicle inspection apparatus. When the wheels of a railway vehicle are attached to the vehicle body, in many cases, a part of the wheels is hidden behind the bogie, so it is difficult to capture the entire surface with one image. Therefore, the present embodiment, which does not remove the wheel from the vehicle body and divides and shoots the wheel while traveling to detect the feature amount, has the effect of greatly reducing the inspection cost.

  Next, a fourth embodiment of the vehicle inspection apparatus will be described. The fourth embodiment has a configuration devised for history management, and will be described with reference to FIGS. FIG. 27 is a block diagram showing a fourth embodiment of the vehicle inspection apparatus, and is a configuration obtained by adding feature amount history display means 13 to the third embodiment.

  The feature amount history display means 13 which is a new configuration will be described. The feature quantity history display means 13 displays the history of the feature quantity stored in the feature quantity history storage means 6. Since the feature amount can be uniquely identified by selecting the wheel ID and feature amount ID from the data of the feature amount history storage means 6 shown in FIG. 11, the same can be obtained by allowing the user to select the wheel ID and feature amount ID. Images taken at different timings are selected from the image database and displayed.

  FIG. 28 shows an example of an input screen for selecting a wheel to be displayed by the feature amount history display means 13. A list of composition IDs is displayed in the list shown in 281. When the user selects an arbitrary composition ID from the list and presses the "Select" button of 283, the composition / wheel is selected using the composition ID selected by the user as a key. The wheel ID attached to the knitting is acquired from the correspondence storage means 5 for each array.

  Further, the latest feature quantity number of the corresponding wheel is searched from the feature quantity history storage means 6 with the acquired wheel ID. The array for each wheel and the feature quantity list acquired as described above are displayed as a list in 284. The user selects a wheel for which the feature history is to be displayed from this list, and presses a “display” button 286. A display example of the history of the selected wheel is shown in FIG.

  In windows 291 and 292 in FIG. 29, images of the selected “wheel 1” taken on different dates are displayed. At this time, the direction of the wheel in the image is displayed by matching the positions of the reference marks 294 and 295, so that the positions of the images taken at different timings can be easily compared. Further, for example, a feature amount desired to be displayed can be displayed on the front as 296, 297 by allowing the user to select a feature amount from a list as shown in 293, for example. In addition, it is possible to adopt an embodiment in which the size of the feature amount is displayed as in 298 and 299, and an enlarged image of the selected feature amount is displayed as in 29a and 29b.

  The embodiment of the feature amount history display means 13 has been described above. However, according to the present invention, it is an object to compare and display the feature amounts of the same wheel that can be collated, and is not limited to the above embodiment.

  In addition, although the method of selecting a wheel for displaying the history is used, an embodiment may be adopted in which a wheel is filtered by the size of a feature amount and a wheel having a feature amount of a certain size or more is displayed as a list.

  As described above, according to the fourth embodiment, the degree of growth of the feature amount can be compared on the image. In the conventional inspection of a stopped vehicle, there was a part hidden by the carriage, and it was not possible to inspect the entire circumference of the wheel without removing the wheel. There is a comparable effect.

  Next, a fifth embodiment of the vehicle inspection apparatus will be described. The fifth embodiment is also a device for effectively performing history management, and manages feature quantities in association with train management operation management information. FIG. 30 is a block diagram showing a fifth embodiment of the vehicle inspection apparatus, which is a configuration obtained by adding a set operation information storage means 14 and a statistical information calculation display means 15 to the third embodiment.

  The new configuration will be described. The train operation information storage means 14 is a storage means for storing information such as the travel distance for each date and the travel section for each train.

  Further, the statistical information calculation and display means includes data on running conditions and feature quantity occurrence statuses of any plural wheels from the knitting operation information storage means 14, the knitting / wheel correspondence storage means 5 and the feature quantity history storage means 6. Is obtained, and a scatter diagram is drawn on a graph, and statistical information is calculated and displayed such as obtaining a correlation value.

  The travel information is stored for each knitting in the knitting operation information storing means 14, but the knitting / wheel correspondence storing means 5 stores the "start date" and "end date" during which the wheels are attached to the knitting. Therefore, it can be acquired by taking over the train operation information for the corresponding period to the wheel.

  As an example of a graph display, plot the “travel distance” and “size of feature” on two axes, or extract only the data of the wheels that have traveled in a certain travel section. It is also possible to plot the feature size generation tendency by plotting the “size of”. Further, paying attention to the position of the wheel during knitting, it is possible to analyze the relationship between the travel distance and the size of the feature amount for each position of the array.

  In this way, according to the fifth embodiment, it is possible to obtain a correlation by seeing a tendency such as the growth of the feature amount not by individual wheels but by the travel distance, travel section, etc. And can be used for preventive maintenance planning.

The above is the fifth embodiment of the vehicle inspection apparatus according to the present invention.
In the present invention, it is important to automatically detect the reference mark and collate the feature values detected on different dates. However, if the mark cannot be detected at all due to the influence of noise or the like, An embodiment is also possible in which a user selects and inputs a position by selecting coordinates from an image, and collates the feature amount position using the position as a reference mark position.

  According to the present invention, vehicle inspection of the wheel side surface is possible.

1: Image input means 2: Feature amount detection means
3: Wheel center detection means
4: Reference mark detection means
5: Knitting / wheel correspondence storage means 6: Feature history storage means
7: Feature amount verification means

Claims (18)

In a vehicle inspection apparatus that detects an image of a wheel side surface of a railway vehicle and detects a position of a feature amount,
Feature amount detection means for detecting a feature amount on a wheel from the input image of the rail vehicle wheel side surface, wheel center detection means for detecting the coordinates of the wheel center from the input image of the rail vehicle wheel side surface, and the input railway Reference mark detection means for detecting a reference mark from an image of a vehicle wheel side surface and determining the coordinates thereof, and a feature for storing the position of the feature amount determined based on the coordinates of the wheel center and the coordinates of the reference mark A vehicle inspection apparatus comprising: a quantity storage unit. In the vehicle inspection device according to claim 1,
The reference mark detection means holds a template for a reference mark, detects an area having a high correlation value by template matching from the input image, and detects the reference mark. The vehicle inspection apparatus according to claim 2,
The template for the reference mark is held as a plurality of templates rotated according to the rotation of the wheel, and a region having a high correlation value is detected by template matching from the input image for each of the plurality of templates. Vehicle inspection device characterized by detecting The vehicle inspection apparatus according to claim 2,
The reference mark detecting means selects a plurality of reference mark candidate areas having a high correlation value by performing matching detection using one type of template, matches the reference mark candidate areas selected by a plurality of templates having different shooting conditions, and obtains a correlation value A vehicle inspection apparatus characterized in that the highest mark position is set as a reference mark position. In the vehicle inspection device according to claim 1,
The reference mark detection means includes character candidate cutout means for cutting out character candidates from the input image, and character arrangement interval rule storage means for storing intervals between characters of a plurality of character strings existing on wheels as rules. A character string candidate that matches the rule is detected from the character string candidates cut out by the character string candidate cut-out means using the rule of the character string arrangement interval stored in the character arrangement interval rule storage means, and the reference mark A vehicle inspection apparatus comprising mark position estimation means for estimating a position. The vehicle inspection apparatus according to claim 5,
The character candidate cutout means of the reference mark detection means estimates the mark position by the mark position estimation means for each of the character string candidates cut out using a plurality of different threshold values, and sets the reference mark position. Inspection device. In the vehicle inspection device according to claim 1,
The reference mark detection unit holds a template for a reference mark, detects a region having a high correlation value by template matching from the input image, and detects a reference mark from the input image. A character candidate cutout unit that cuts out character candidates, a character arrangement interval rule storage unit that stores, as a rule, an interval between characters of a plurality of character strings existing on the wheel, and a character string stored in the character arrangement interval rule storage unit A mark position estimating means for detecting a character string candidate that matches the rule from the character string candidates cut out by the character string candidate cut-out means, the reference mark matching means, and the mark position estimation. Mark position determining means for finally determining the mark position from the mark position candidates obtained by the means Vehicle inspection device, characterized in that. The vehicle inspection apparatus according to claim 7,
The mark position determining unit determines a reference mark position in preference to the reference mark position candidate estimated by the mark position determining unit when the correlation value of the reference mark position candidate detected by the reference mark matching unit is sufficiently high. Vehicle inspection device characterized by the above. The vehicle inspection apparatus according to claim 7,
When there are a plurality of reference mark position candidates detected by the reference mark matching means and a plurality of reference mark position candidates estimated by the mark position determining means, the mark position determining means detects the same position or the number of cases estimated A vehicle inspection apparatus characterized in that a large number of positions are preferentially determined as reference mark positions. The vehicle inspection apparatus according to claim 7,
A plurality of different rules stored in the character arrangement interval rule storage means are weighted, and based on the weighting, a priority is determined when the mark position determination means determines a reference mark position from a plurality of reference mark position candidates. A vehicle inspection device. The vehicle inspection apparatus according to claim 7,
When there are a plurality of reference mark position candidates detected by the reference mark matching means, and a plurality of reference mark position candidates estimated by the mark position determination means, the mark position determination means, when there are a plurality of estimated positions in the vicinity, A vehicle inspection apparatus characterized in that when a plurality of estimated positions are averaged or a plurality of rules are weighted, a position calculated by taking the weighted average is determined as a mark position. In the vehicle inspection device that processes the image of the side of the wheel of the railway vehicle and records the history of the feature amount,
Feature amount detection means for detecting a feature amount on a wheel from the input image of the rail vehicle wheel side surface, wheel center detection means for detecting the coordinates of the wheel center from the input image of the rail vehicle wheel side surface, and the input railway A reference mark detecting means for detecting a reference mark from an image of a vehicle wheel side surface and determining its coordinates; a knitting / wheel correspondence storing means for storing correspondence of wheel identification information for identifying a photographing wheel when photographing a railway vehicle; Obtained from feature image history storage means for storing the position of the feature value determined based on the coordinates of the wheel center and the coordinates of the reference mark in association with the wheel identification information, and the inputted image of the side surface of the railroad vehicle wheel. If there is a feature amount that is considered to be the same from the feature amount position information, the information about the wheel having the same wheel identification information is obtained from the feature amount history storage means, Vehicle inspection apparatus comprising: a feature quantity matching means to append the feature value history storing means to impart identification information indicating a feature amount. The vehicle inspection apparatus according to claim 12,
A vehicle inspection apparatus comprising a feature amount history display means for displaying images of the same wheels taken on different dates side by side so that the same feature amount can be compared. The vehicle inspection apparatus according to claim 12,
A vehicle inspection apparatus comprising: a train operation information storage unit; and a statistical information calculation display unit, wherein the relationship between information relating to traveling of wheels and information relating to feature quantities is displayed in comparison. The vehicle inspection apparatus according to claim 12,
When the reference mark cannot be detected by the reference mark detection unit and the mark position determination unit, the reference mark detection unit includes a reference mark position input unit that inputs the position of the reference mark on the image. Based on the input reference mark position, A vehicle inspection apparatus characterized by collating a quantity position and storing it in a feature quantity history storage means. In a vehicle inspection device that obtains a continuous image for one lap of a wheel from multiple images taken at multiple locations close to the wheel side of a railway vehicle, and detects a feature amount,
Feature amount detection means for detecting feature values on wheels for a plurality of images of the input railcar wheel side surface, wheel center detection means for detecting coordinates of a wheel center for the plurality of images of the input railcar wheel side surface, A reference mark detecting means for detecting a reference mark for a plurality of images on the side surface of the input railcar wheel and determining coordinates thereof, and an inter-image position matching means for obtaining a corresponding position for images taken at adjacent positions and connecting them. A vehicle inspection apparatus comprising: The vehicle inspection device according to claim 16,
The inter-image position checking means obtains the amount of wear of the wheel by detecting the center of the wheel on the image and the arc of the wheel, and calculates a corresponding point between adjacent images. The vehicle inspection device according to claim 16,
A vehicle inspection device that detects a feature amount on a wheel from an image obtained by polar-coordinate conversion using the detected wheel center as a reference.

Images