TWI326049B - Method of image object classification and identification - Google Patents

Description

达达编号·· TW3286PA IX. Description of the Invention: [Technical Field] The present invention relates to an image processing method, and more particularly to a method for classifying and identifying an image object. [Prior Art] The motion detection function is one of the basic functions of the current surveillance video recording system. In order to reduce the amount of data stored, the conventional security video recording system only starts the recording function or issues an alarm when it detects the movement of the object. Despite this, the current surveillance video recording system often has misjudgment of motion detection. For example, if the lens image is shaken by the wind, or the environmental factors such as light changes, the money will be recorded or sent out. Alerts, which in turn lead to waste of data storage and false alarms. Therefore, if the security surveillance video system can be combined with image processing technology, the input === line classification and identification 'identify that the mobile object belongs to the second == car, luggage, etc.), will be only for the target that really wants to be viewed Alerts are issued and the chances of false alarms are greatly reduced. SUMMARY OF THE INVENTION It is to be noted that the purpose of the present invention is to classify and identify methods, and to obtain fast & When applied to the image monitoring system, the object classification and identification alarms occur, and the errors can be effectively and surely reduced. According to the object of the present invention, one-to-be-like stock is reduced. Image object identification method, package

132 (4) Dirty: TW3286PA image curve drawing - object release object rim (four) rim wave type 'converted object wheel wave shape from space domain to frequency 吏.:: contour wave type and - wave type database library search Reference object wheel Yi H and (Beacon reference object contour wave type corresponds to the reference object. The piece is the following step object classification method, - to consume the new # according to the reference wheel • curve edge system, the next month The above objects, features, and advantages can be more clearly illustrated as a preferred embodiment of the present invention, and in conjunction with the accompanying drawings, in detail. ~ The cow: The flow chart of the identification method. The image object recognition method starts with taking an image object m, and draws an object contour wave pattern according to the contour curve of the image in step (4) G. The preferred edge mode of the object contour wave pattern is as follows. Figure 5A to 5C. As shown in Figure 5A, image 6 1326049

Nickname: TW3280PA The object's rounded curve has several point coordinates (Xi, Yi). First, the center of gravity coordinates (Xe, Ye) of the image object are shown. The center of gravity coordinate of the image object, including XC, YC, is the sum of the coordinates of each point (Xi, Yi) on the rim curve. Furthermore, 'calculate the coordinates of each point coordinate (xi, Yi) and the center of gravity * (^^'丫〇 on the round therapy curve. You can calculate from the point on the contour curve = ' For example: the starting point of Figure 5B (10) Shoulder) is the first point encountered when scanning the object wheel X-axis direction and then sweeping the γ-axis direction. = Point calculates the distance between each point and the center of gravity along the corridor curve. The contour coordinates of the point coordinates (Xl, Yi) and the center of gravity coordinates (Xc, Yc). The 5th number, and its ordinate is the coordinate map of the = type, the abscissa is the wheel point step _ the original object contour wave shape in the 5C picture is normal inch, and the second M scales the original object round wave type To - fixed size, and become a positive boundary of the 5D map + Zha from the case of zooming in or out to the case of 'the object is equal. Step _ further transforms the == object size not-profile type wave (four), and the object wheel =:: changes to the wave of the 1 rate domain ==: empty: MW conversion, or wavelet conversion. Γ-type via waveform conversion can exclude the opposite of the image object rotating or mirroring 7

1326049 : Sanda number: TW3286PA 〗 Before, type pair, you can filter the object in the step of the first step to filter the object wave wave S-frequency boring tool using low-pass chopper to filter the high-frequency part of the contour wave of the object, regret low Solution. In this way, the high-frequency vibration I in the wheel-type type caused by the integrity of the object can be removed, and the rut is smoothed. Then, in step 150*, the contour waveform and the contour database of the object after the normal J are compared, and the contour waveform of the object is extracted from the waveform. The wave type database has several profiles, which can be contour waveforms of objects such as people, cars and luggage. Λ >, Fig. 2, which shows a flow chart of the comparison of the sill and the library in the image object identification method of Fig. 1. First, in step 251+, a reference object contour waveform is selected from the waveform data, and in step 252, the selected object surface waveform is matched with the positive and the contour contour wave / : to obtain a wave pattern difference. Next, in the step of the towel, the value of the wave + Sao bar value or the sum of the least squares is calculated and defined as an alignment value. At 4, the _ comparison value is the minimum and less than the - threshold value. If the next % goes to the next step; if not, then returns to step 251, select another reference object contour waveform from the waveform 2, 52, and repeat the step small comparison step, 253 and step 254 until It is judged that the second round of production is less than the threshold value. At this point, there are different specific parts due to different object types. Type, so it can be judged by the contour wave type which object is the image object and can directly enter the final step 160, and it is judged that the image object should be referenced by the reference object wheel type with the smallest comparison value. One pair: identify the pieces. The method may further perform step 255 and step 256, so that the auxiliary s w image object system is the corresponding reference object. Step 2 5 5 Judgment 8

No.: TW3286PA 1326049 Whether the aspect ratio (height/widthratio) of the object is within a certain range, and step 256 determines whether the color histogram of the image object conforms to the color histogram of the corresponding reference object. Figure. In addition, the way in which the above-mentioned waveform database is established is adopted once the classification method. Please refer to Fig. 3, which is a flow chart of a method for classifying image objects according to an example. The image object recognition method is a reference object, and in step 310, a reference object contour waveform is drawn according to the - reference line. Reference = 3 = good drawing method is also the same as the center of gravity coordinates (Xc, k) of the image of the image of the rotation of the image of the test (4); the coordinates of each point on the rim curve (xi ° 2 and the number of turns is _ , __=== object contour map. Then, the cow wq 9 ',,, will be the first step of the sninAA u疋 size. Step 330 is replaced by a change, so that the regular compound / round wave type wave mode to - The type of the frequency domain is transformed from the waveform of the spatial domain to the object of the temple, etc. Finally, the normalization and classification method of the image object. The wave type is used to end the rim type of the object object. All are classified by this method and then stored in the waveform database +. 9 1J26049

i face number: TW3286PA Please refer to Figure 4, 1 笸 笸 笸. In step 4 () 1, the ^object subtraction divides the moving object into a background plane. Then, the image of the image is stored in the image of the enemy in the step 402. 庇 庇 庇 〆 〆 〆 〆 〆 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : The image object to be recognized is also referred to as (4) the reference object of the touch method. π is used in the image classification method to be classified as the above-mentioned real image of the present invention (the fourth), using the shape of the wheel of the image object: Fang: The method of wave-type money to a fixed value can be excluded = the image of the iu彡 image created by the lens is the same as that of the lens; the conversion of the brother-two wave type to the frequency domain wave type can be discharged positively. The situation caused by the rotation of the image object and the exclusion of the framing finder == the opposite of the mirror caused by the difference in direction. Therefore, the method of the above-described method of the present invention can (4) quickly and accurately classify objects and identify the results. In addition, when the present invention is applied to an image monitoring system, it is possible to quickly determine which object the detected moving object is by recognizing the contour waveform of the moving object, and only issue an alarm for the target that is actually intended to be monitored. , and greatly reduce the chance of false alarms, and can effectively and surely reduce the amount of image data storage.

达 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 It will be apparent to those skilled in the art that various changes and modifications can be made without departing from the spirit and scope of the invention. Therefore, the scope of the invention is defined by the scope of the appended claims.

11 1326049

3: TW3286PA / [Simple Description of the Drawings] FIG. 1 is a flow chart showing an image object identification method according to a preferred embodiment of the present invention. Fig. 2 is a flow chart showing the comparison of the data bases in the image object identification method of Fig. 1. Figure 3 is a flow chart showing an image object classification method in accordance with a preferred embodiment of the present invention. FIG. 4 is a flow chart showing image capture in the image object identification method of FIG. 1 and the image object classification method of FIG. ^ Figures 5A to 5E are schematic views showing the contour waveform of the formed object. [Main component symbol description] (none)

12

Claims (1)

1326042 : up to: TW3286PA X. Patent application scope · · 1 · An image object identification method, including: (a) according to one of the image object wheel shape; curve drawing an object contour wave object object contour wave shape, to scale the Switching to the traveling wave type conversion, the contour waveform of the object is converted from the spatial domain and the converted contour wave pattern and the -wave type of the object are searched for a reference object contour waveform; Refer to the object! Broken the shirt image object is one of the reference object rim wave type correspondence method, please refer to the image object identification method described in the patent '1 item'. The eight steps (a) include: Secret = one of the image objects Center of gravity coordinates, where the center of gravity coordinate is the sum of the coordinates of each point on the line of ~1卩曲~; and calculating the distance between the coordinates of each point on the contour curve and the coordinates of the center of gravity' and the object shape of the object according to μ . The image object recognition method of claim 1, wherein before the step (d), the method further comprises: (f) establishing the waveform database. 4. The image object identification method according to item 3 of the patent application scope, wherein the step (f) includes: 13 violation number: TW3286PA 1326049 8. The image object identification method according to claim 1, wherein The step (d) includes: determining whether the comparison value of the reference object wheel-wave type and the normalized and converted object contour waveform is less than a threshold value. 9. The image object identification method according to claim 8, wherein the step (d) comprises: dividing the plurality of reference object contour waveforms of the waveform database with the normalized (4) object Contour wave type subtraction, wave type difference; τι double like, calculate the absolute value or the sum of the least squares of the wave type differences, and is a plurality of the comparison values; and select one of the comparison values The comparison value 'and the minimum alignment value is less than a threshold value. 10. The image object identification method of claim 8, wherein the step (d) further comprises: determining whether the aspect ratio of the image object is within a specific range, to assist in identifying the image object as the corresponding Reference object. 11. The image object identification method according to claim 8, wherein the step (d) further comprises: 11 determining whether the color histogram of the image object conforms to the color histogram of the corresponding reference object to assist in identifying the image. The object is the corresponding reference object. 12. The image object identification method according to claim 1, wherein the waveform conversion of the step (C) is performed by Fourier transform, z-turn 15 1^^0049 * two-number: TW3286PA, cosine transform, Or wavelet conversion. 13. The image object identification method according to claim 1, wherein before the step (a), the method further comprises: storing the image surface without the moving object as a background surface; capturing an image monitoring system And an image of the image; and the image of the image is separated from the background surface to separate the displaced image object. 14. The image object recognition method according to claim 1, wherein before the step (d), the method further comprises: filtering the high frequency portion of the contour waveform of the object. 15. An image object classification method comprising: (a) drawing a reference object wheel grinding pattern according to a reference contour curve of a reference object; (b) normalizing the reference object contour wave plow to scale the reference object contour Waveform to a fixed size; (c) performing a mode conversion such that the reference object contour is converted from the spatial φ domain to the frequency domain; and (d) storing the normalized and converted reference object in a wave-wave type In a wave type database. The image object classification method according to claim 15, wherein the step (a) comprises: calculating a reference gravity center coordinate of the reference image object, wherein the reference weight ~ coordinate system is on the reference contour curve The sum of the coordinates of each reference point coordinate; and the number: TW3286PA 1326049 Calculate the distance between each reference point coordinate on the reference contour and the reference centroid coordinate, and draw the reference object contour waveform accordingly. The image object classification method according to claim 15, wherein the waveform conversion of the step (c) is Fourier transform, Z transform, cosine transform, or wavelet transform. 18. The image object classification method according to claim 15, wherein before the step (a), the method further comprises: storing the image surface without the moving object as a background image; capturing an image monitoring system The image is scanned; and the captured image is subtracted from the background surface to separate the reference object of the displacement. 17