CN108875577A - Object detection method, device and computer readable storage medium - Google Patents
Object detection method, device and computer readable storage medium Download PDFInfo
- Publication number
- CN108875577A CN108875577A CN201810453480.5A CN201810453480A CN108875577A CN 108875577 A CN108875577 A CN 108875577A CN 201810453480 A CN201810453480 A CN 201810453480A CN 108875577 A CN108875577 A CN 108875577A
- Authority
- CN
- China
- Prior art keywords
- anchor frame
- coordinate
- prediction block
- frame
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of object detection methods, include the following steps：Image to be detected is obtained, described image to be detected extracts feature by multilayer convolution in neural network and generates characteristic pattern；Modified structural parameters in neural network model are loaded, generate corresponding anchor frame coordinate based on the structural parameters, wherein the preset structure parameter includes the length-width ratio of the reference dimension of anchor frame, anchor frame scale and anchor frame；Candidate frame coordinate is generated based on region nomination subnet, takes corresponding region to obtain individual features by area-of-interest pond according to candidate frame coordinate on characteristic pattern；Prediction block coordinate is determined based on the feature, and target object location is determined based on the prediction block coordinate.The invention also discloses a kind of object detecting device and computer readable storage mediums.Prediction block determines target object after the present invention realizes generation optimization, can detect to compared with Small object, improves the verification and measurement ratio of target.
Description
Technical field
The present invention relates to technical field of image detection more particularly to a kind of object detection methods, device and computer-readable
Storage medium.
Background technique
The detection of target is widely used in every field in image, for example, traffic refers in image in automatic Pilot field
Show that the detection of board is very important link, the purpose is to the traffic sign positions in detection image, and then refer to by traffic
Show the identification of board, the traveling of guiding vehicle guarantees traffic safety.
Currently, in detection image when the lesser target of area, feature letter in the characteristic pattern that is obtained due to feature extractor
It ceases considerably less, it means that be difficult to be positioned and be classified, for example, detector is in detection as the Small object of traffic sign etc
When it is extremely difficult, so, current object detection method is difficult to detect Small object.
Above content is only used to facilitate the understanding of the technical scheme, and is not represented and is recognized that above content is existing skill
Art.
Summary of the invention
The main purpose of the present invention is to provide a kind of object detection method, device and computer-readable mediums, it is intended to solve
Certainly current detection method is difficult to the technical issues of detecting to Small object.
To achieve the above object, the present invention provides a kind of object detection method, and the object detection method includes following step
Suddenly：
Image to be detected is obtained, described image to be detected extracts feature by multilayer convolution in neural network and generates feature
Figure；
Modified structural parameters in neural network model are loaded, corresponding anchor frame is generated based on the structural parameters and sits
Mark, wherein the preset structure parameter includes the length-width ratio of the reference dimension of anchor frame, anchor frame scale and anchor frame；
Candidate frame coordinate is generated based on region nomination subnet, takes corresponding region to pass through according to candidate frame coordinate on characteristic pattern
It crosses area-of-interest pond and obtains individual features；
Prediction block coordinate is determined based on the feature, and target object location is determined based on the prediction block coordinate.
Preferably, described the step of generating anchor frame coordinate based on the structural parameters, includes：
Area based on the reference dimension and the anchor frame dimension calculation anchor frame；
The length and width of anchor frame are calculated, based on the area and anchor frame length-width ratio to determine the size of anchor frame；
The centre coordinate of anchor frame is obtained, and determines the position of anchor frame based on the centre coordinate, generates anchor frame four edges
Coordinate.
Preferably, it is based on after the step of structural parameters generate corresponding anchor frame coordinate, the target detection side
Method further includes：
Determine whether the anchor frame is more than original image range；
When the anchor frame is more than the original image range, the anchor frame more than original image range is eliminated, to obtain target anchor frame seat
Mark.
Preferably, described that prediction block coordinate is determined based on the feature, and object is determined based on the prediction block coordinate
The step of body position includes：
The feature obtains offset further across recurrence；
Final prediction block coordinate is obtained based on the corresponding target anchor frame coordinate of the offset correction, to determine object
Body position.
It is preferably, described that final prediction block coordinate is obtained based on the corresponding target anchor frame coordinate of the offset correction,
After the step of determining target object location, the object detection method further includes：
Duplicate removal processing is carried out to the prediction block using preset algorithm, to obtain target prediction frame.
Preferably, described that duplicate removal processing is carried out to the prediction block using preset algorithm, to obtain the step of target prediction frame
Suddenly include：
The confidence level of the prediction block is obtained, and is ranked up the prediction block based on the confidence level, to be arranged
Sequence result；
It is successively chosen based on the ranking results and hands over and compare when the highest prediction block of previous belief is calculated with other prediction blocks；
Determine the friendship and than whether being greater than second threshold；
The friendship and the prediction block than being greater than the second threshold are eliminated, retains the friendship and than no more than second threshold
Target prediction frame.
Preferably, the neural network includes the extractor of target object character pair, candidate frame generation network, candidate frame
Region Feature Extraction, classification and Recurrent networks.
In addition, to achieve the above object, the present invention also provides a kind of object detecting device, object detecting device includes：It deposits
Reservoir, processor and it is stored in the object detection program that can be run on the memory and on the processor, the target
The step of detection program realizes any of the above-described object detection method when being executed by the processor.
In addition, to achieve the above object, it is described computer-readable the present invention also provides a kind of computer readable storage medium
It is stored with object detection program on storage medium, any of the above-described target is realized when the object detection program is executed by processor
The step of detection method.
The present invention extracts feature by multilayer convolution in neural network by obtaining image to be detected, described image to be detected
Characteristic pattern is generated, modified structural parameters in neural network model are then loaded, is generated based on the structural parameters corresponding
Anchor frame coordinate, wherein the preset structure parameter includes the length-width ratio of the reference dimension of anchor frame, anchor frame scale and anchor frame, then
Candidate frame coordinate is generated based on region nomination subnet, takes corresponding region by interested according to candidate frame coordinate on characteristic pattern
Pool area obtains individual features, finally determines prediction block coordinate based on the feature, and determine based on the prediction block coordinate
Target object location；It is thus achieved that prediction block determines target object after generating optimization, can be detected to compared with Small object,
Improve the verification and measurement ratio of target.
Detailed description of the invention
Fig. 1 is that the structure of the affiliated terminal of object detecting device in hardware running environment that the embodiment of the present invention is related to is shown
It is intended to；
Fig. 2 is the flow diagram of object detection method first embodiment of the present invention；
Fig. 3 is the schematic diagram for the anchor frame that the present invention generates；
Fig. 4 is to generate anchor frame coordinate step based on the structural parameters described in object detection method second embodiment of the present invention
Rapid refinement flow diagram；
Fig. 5 is the flow diagram of object detection method 3rd embodiment of the present invention；
Fig. 6 is to be sat in object detection method fourth embodiment of the present invention based on the corresponding target anchor frame of the offset correction
Mark obtains final prediction block coordinate, to determine the flow diagram of target object location step；
Fig. 7 is the flow diagram of the 5th embodiment of subject invention detection method；
Fig. 8 is to be carried out at duplicate removal using preset algorithm to the prediction block in object detection method sixth embodiment of the present invention
Reason, to obtain the flow diagram of target prediction frame step.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
As shown in Figure 1, Fig. 1 be in hardware running environment that the embodiment of the present invention is related to belonging to object detecting device eventually
The structural schematic diagram at end.
The terminal of that embodiment of the invention can be PC.As shown in Figure 1, the terminal may include：Processor 1001, such as CPU,
Network interface 1004, user interface 1003, memory 1005, communication bus 1002.Wherein, communication bus 1002 is for realizing this
Connection communication between a little components.User interface 1003 may include display screen (Display), input unit such as keyboard
(Keyboard), optional user interface 1003 can also include standard wireline interface and wireless interface.Network interface 1004 is optional
May include standard wireline interface and wireless interface (such as WI-FI interface).Memory 1005 can be high speed RAM memory,
It is also possible to stable memory (non-volatile memory), such as magnetic disk storage.Memory 1005 optionally may be used also
To be independently of the storage device of aforementioned processor 1001.
Optionally, terminal can also include camera, RF (Radio Frequency, radio frequency) circuit, sensor, audio
Circuit, WiFi module etc..Wherein, sensor such as optical sensor, motion sensor and other sensors.Specifically, light
Sensor may include ambient light sensor and proximity sensor, wherein ambient light sensor can according to the light and shade of ambient light come
The brightness of display screen is adjusted, proximity sensor can close display screen and/or backlight when mobile terminal is moved in one's ear.As
One kind of motion sensor, gravity accelerometer can detect the size of (generally three axis) acceleration on direction, when static
Size and the direction that can detect that gravity can be used to identify application (such as the horizontal/vertical screen switching, related trip of mobile terminal posture
Play, magnetometer pose calibrating), Vibration identification correlation function (such as pedometer, tap) etc.；Certainly, mobile terminal can also configure
The other sensors such as gyroscope, barometer, hygrometer, thermometer, infrared sensor, details are not described herein.
It will be understood by those skilled in the art that the restriction of the not structure paired terminal of terminal structure shown in Fig. 1, can wrap
It includes than illustrating more or fewer components, perhaps combines certain components or different component layouts.
As shown in Figure 1, as may include operation server, network in a kind of memory 1005 of computer storage medium
Communication module, Subscriber Interface Module SIM and object detection program.
In terminal shown in Fig. 1, network interface 1004 is mainly used for connecting background server, carries out with background server
Data communication；User interface 1003 is mainly used for connecting client (user terminal), carries out data communication with client；And processor
1001 can be used for calling the object detection program stored in memory 1005.
In the present embodiment, object detecting device includes：Memory 1005, processor 1001 and it is stored in the memory
On 1005 and the object detection program that can be run on the processor 1001, wherein processor 1001 calls memory 1005
When the object detection program of middle storage, following operation is executed：
Image to be detected is obtained, described image to be detected extracts feature by multilayer convolution in neural network and generates feature
Figure；
Modified structural parameters in neural network model are loaded, corresponding anchor frame is generated based on the structural parameters and sits
Mark, wherein the preset structure parameter includes the length-width ratio of the reference dimension of anchor frame, anchor frame scale and anchor frame；
Candidate frame coordinate is generated based on region nomination subnet, takes corresponding region to pass through according to candidate frame coordinate on characteristic pattern
It crosses area-of-interest pond and obtains individual features；
Prediction block coordinate is determined based on the feature, and target object location is determined based on the prediction block coordinate.
Further, processor 1001 can call the object detection program stored in memory 1005, also execute following
Operation：
Area based on the reference dimension and the anchor frame dimension calculation anchor frame；
The length and width of anchor frame are calculated, based on the area and anchor frame length-width ratio to determine the size of anchor frame；
The centre coordinate of anchor frame is obtained, and determines the position of anchor frame based on the centre coordinate, generates anchor frame four edges
Coordinate.
Further, processor 1001 can call the object detection program stored in memory 1005, also execute following
Operation：
Determine whether the anchor frame is more than original image range；
When the anchor frame is more than the original image range, the anchor frame more than original image range is eliminated, to obtain target anchor frame seat
Mark.
Further, processor 1001 can call the object detection program stored in memory 1005, also execute following
Operation：
The feature obtains offset further across recurrence；
Final prediction block coordinate is obtained based on the corresponding target anchor frame coordinate of the offset correction, to determine object
Body position.
Further, processor 1001 can call the object detection program stored in memory 1005, also execute following
Operation：
Duplicate removal processing is carried out to the prediction block using preset algorithm, to obtain target prediction frame.
Further, processor 1001 can call the object detection program stored in memory 1005, also execute following
Operation：
The confidence level of the prediction block is obtained, and is ranked up the prediction block based on the confidence level, to be arranged
Sequence result；
It is successively chosen based on the ranking results and hands over and compare when the highest prediction block of previous belief is calculated with other prediction blocks；
Determine the friendship and than whether being greater than second threshold；
The friendship and the prediction block than being greater than the second threshold are eliminated, retains the friendship and than no more than second threshold
Target prediction frame.
The present invention further provides a kind of object detection methods.It is object detection method first of the present invention referring to Fig. 2, Fig. 2
The flow diagram of embodiment.
In the present embodiment, which includes the following steps：
Step S10 obtains image to be detected, obtains image to be detected, described image to be detected is by more in neural network
Layer convolution extracts feature and generates characteristic pattern；
In the present embodiment, which extracts feature by multilayer convolution in network and generates characteristic pattern for generation
Characteristic pattern is sent into region nomination subnet and generates candidate frame coordinate, is then taken on the characteristic pattern of generation according to candidate frame coordinate corresponding
Region carry out area-of-interest pond, and then be sent into classifier classify.
Further, which refers to the image for the target object that needs detect, which can be electronics and set
The image or electronic equipment is sent to by acquisition equipment acquired image that standby local preservation perhaps caches, for example, camera
The image of collected traffic sign is sent to electronic equipment.The electronic equipment includes mobile phone, plate, computer etc..Certainly,
According to trained neural network model, image to be detected carries out convolution algorithm by multilayer convolution and extracts corresponding characteristic pattern,
This feature figure is zoomed in and out according to certain proportion, for example, by the characteristic pattern extracted narrow down to original image size 16/
One.
Step S20 loads modified structural parameters in neural network model, is generated based on the structural parameters corresponding
Anchor frame coordinate, wherein the preset structure parameter includes the length-width ratio of the reference dimension of anchor frame, anchor frame scale and anchor frame；
In the present embodiment, corresponding region, the quantity of anchor frame are determined according to the coordinate of anchor frame, candidate frame, prediction block
It can be 1,000,2,000 etc., anchor frame can regard what the candidate region that target object is likely to occur in the picture was formed as
Rectangle frame, referring to Fig. 3, Fig. 3 is the schematic diagram that anchor frame generates, and can set different reference area, according to reference area according to
Different length-width ratios generates different anchor frames, and anchor frame can be rectangle frame, for example, three kinds of areas 1282,2562,5122 are pressed respectively
According to three kinds of ratios 1:1,1:2,2:1 can be generated nine rectangle frames, generate of course, it is possible to set a variety of areas according to a variety of ratios
Rectangle frame, for example, the area of setting can also be 3842,6402 etc., the Aspect Ratio of setting can be 2:3,3:1 etc..
Further, the preset structure parameter in neural network model can also be obtained, which includes anchor
The scale and length-width ratio of the reference dimension of frame, anchor frame determine the size and location of anchor frame according to preset structure parameter, generate anchor
Frame eliminates the anchor frame then more than characteristic pattern range, retains the target anchor frame for being not above characteristic pattern range, utilizes neural network
Extract the candidate frame in the target anchor frame.
Further, to the base of preset anchor frame in two stages target detection convolutional neural networks region nomination sub-network structures
The structural parameters such as the length-width ratio of object staff cun, scale and anchor frame determine the size and number of generation anchor frame, and the pixel in image is determined
Determine the position of anchor frame, which is configured by technical staff, wherein reduce the reference dimension of anchor frame, increase anchor
When the scale of frame, then some smaller anchor frames will be generated, the anchor frame of these small sizes is then the basis for detecting Small object.
Step S30 generates candidate frame coordinate based on region nomination subnet, is taken accordingly on characteristic pattern according to candidate frame coordinate
Region obtain individual features by area-of-interest pond；
Step S40 determines prediction block coordinate based on the feature, and determines object position based on the prediction block coordinate
It sets.
In the present embodiment, generating a depth after the convolutional layer effect that this region is 3x3 by a core size is 256
Characteristic pattern, this characteristic pattern pass through respectively a core size be 1x1 convolutional layer, coordinate return layer and classification layer obtain 4k sit
Mark and 2k score, wherein k is the quantity of anchor frame, and former anchor frame quantity is 9=3 (scales) x3 (aspect ratios), to knot
Structure parameter obtains modified structural parameters after being made that modification, has used more scales (scales) and ratio (aspect
Ratios), more anchor frame coordinates are obtained, will be mapped to the corresponding coordinate of characteristic pattern using pixel in original image as center coordinate
In original image, the position of anchor frame is determined according to the calculated anchor frame size of structural parameters and centre coordinate.
Further, it is determined that the target object includes the position of the classification of target object and the target object in determining image
It sets, position includes coordinate.Neural network model can export the corresponding confidence level of each detection block, to the confidence of each prediction block
Degree according to being ranked up from big to small, to obtain ranking results, is then based on the ranking results and successively chooses and work as previous belief most
High prediction block calculates with other prediction blocks and hands over and compare, and eliminates friendship and the prediction block than being unsatisfactory for preset condition, retains and hands over and compare
Meet the prediction block of preset condition.When being trained, the classification of the target object in image can be labeled, further include
The callout box coordinate of object is labeled, can be calculated according to the corresponding coordinate of prediction block and the corresponding coordinate of callout box
It hands over and compares.
The object detection method that the present embodiment proposes, by obtaining image to be detected, described image to be detected is by nerve
Multilayer convolution extracts feature and generates characteristic pattern in network, then loads modified structural parameters in neural network model, is based on
The structural parameters generate corresponding anchor frame coordinate, wherein the preset structure parameter includes the reference dimension of anchor frame, anchor frame ruler
The length-width ratio of degree and anchor frame then generates candidate frame coordinate based on region nomination subnet, according to candidate frame coordinate on characteristic pattern
It takes corresponding region to obtain individual features by area-of-interest pond, prediction block coordinate is finally determined based on the feature, and
Target object location is determined based on the prediction block coordinate；It realizes and generates more anchor frames seats according to modified structural parameters
Mark, and determine that prediction block coordinate can detect Small object, improve the verification and measurement ratio of target.
Based on first embodiment, the second embodiment of object detection method of the present invention is proposed, reference Fig. 4, in the present embodiment,
Step S20 includes：
Step S21, the area based on the reference dimension and the anchor frame dimension calculation anchor frame；
Step S22 calculates the length and width of anchor frame, based on the area and anchor frame length-width ratio to determine the size of anchor frame；
Step S23 obtains the centre coordinate of anchor frame, and the position of anchor frame is determined based on the centre coordinate, generates anchor frame
The coordinate of four edges.
In the present embodiment, the size that can determine anchor frame according to the reference dimension of anchor frame, scale and length-width ratio, according to figure
The pixel of pixel can determine the position of anchor frame as in, using pixel as center Coordinate generation anchor frame.According to the benchmark of anchor frame
The scale of size and anchor frame can calculate the area of anchor frame, and the length and width of anchor frame can be calculated further according to the length-width ratio of anchor frame, from
And determine the size of anchor frame, for example, the reference dimension of anchor frame is 16, the scale of anchor frame is (2,4,8), and length-width ratio is (1:1,2:
1,1:2), three scales, three length-width ratios can determine 3x3=9 anchor frame, choose one of anchor frame and carry out example, station meter
Very little is 16, scale 4, length-width ratio 2:1, then the area S of this anchor frame is then (16*4)2=4096, it can according to length-width ratio
In the hope of lengthIt is wide((S=H*W=4096) then chooses coordinate centered on a pixel again,
It can then determine position and the size of the anchor frame.
The object detection method that the present embodiment proposes, by being based on the reference dimension and the anchor frame dimension calculation anchor frame
Area, be then based on the area and anchor frame length-width ratio and calculate the length of anchor frame and wide, to determine the size of anchor frame, finally obtain
The centre coordinate of anchor frame, and determine based on the centre coordinate position of anchor frame, generate the coordinate of anchor frame four edges；Realize root
The size and location of anchor frame are determined according to structural parameters and centre coordinate, to generate anchor frame coordinate.
Based on first embodiment, the 3rd embodiment of object detection method of the present invention is proposed, reference Fig. 5, in the present embodiment,
After step S20, further include：
Step S50 determines whether the anchor frame is more than original image range；
Step S60 eliminates the anchor frame more than original image range, when the anchor frame is more than the original image range to obtain mesh
Mark anchor frame coordinate.
In the present embodiment, preset according to neural network due to being center coordinate according to pixel each on characteristic pattern
Structural parameters generate anchor frame, some anchor frames have exceeded the range of characteristic pattern in the anchor frame of generation, so to eliminate more than characteristic pattern
Then range anchor frame passes through several layers of structures to obtain the target anchor frame for being not above characteristic pattern range inside neural network
Processing eliminate and can not have the anchor frame of target object in target frame, there may be target objects in the anchor frame stayed, will
The anchor frame stayed as target anchor frame, by the coordinate to target anchor frame.
Whether the object detection method that the present embodiment proposes, be more than original image range by the determination anchor frame, then work as institute
When stating anchor frame more than the original image range, the anchor frame more than original image range is eliminated, to obtain target anchor frame coordinate；Realize elimination
The anchor frame of redundancy, to retain, there may be the target anchor frames of target object, and then improve the verification and measurement ratio of target.
Based on 3rd embodiment, the fourth embodiment of object detection method of the present invention is proposed, reference Fig. 6, in the present embodiment,
Step S40 includes：
Step S41, the feature obtain offset further across recurrence；
Step S42 obtains final prediction block coordinate based on the corresponding target anchor frame coordinate of the offset correction, with true
Set the goal object space.
In the present embodiment, subnet is nominated according to region and generates candidate frame coordinate, according to candidate frame coordinate on characteristic pattern
Corresponding region is taken to obtain individual features by area-of-interest pond, this feature obtains offset further across recurrence, should
Offset obtains final prediction block coordinate for correcting corresponding anchor frame coordinate, and each group of offset is used to correct and candidate frame
Concentric anchor frame.For example, the coordinate record in the upper left corner of anchor frame and the lower right corner is got off, it is denoted as { x0, y0, x1, y1 }, if
Offset is {+1, -1 }, then { x0+1, y0-1, x1+1, y1-1 } can be denoted as by obtaining prediction block after deviating, then according to candidate
The centre coordinate of frame obtains the coordinate of prediction block.
The object detection method that the present embodiment proposes, obtains offset further across recurrence by the feature, then
Final prediction block coordinate is obtained based on the corresponding target anchor frame coordinate of the offset correction, to determine target object location；
It realizes and prediction block coordinate is obtained according to offset, to further determine that target object position, improve the inspection of target
Survey rate.
Based on fourth embodiment, the 5th embodiment of object detection method of the present invention is proposed, reference Fig. 7, in the present embodiment,
After step S42, further include：
Step S43 carries out duplicate removal processing to the prediction block using preset algorithm, to obtain target prediction frame.
In the present embodiment, which includes non-maxima suppression method, when obtaining prediction block, using it is non-greatly
It is worth inhibition method and duplicate removal processing is carried out to prediction block, calculates each prediction first with the sorter network in neural network model first
The confidence level of frame, then being ranked up according to sequence from big to small to confidence level, calculates prediction block according to ordering rule
It hands over and compares, so that removal is unsatisfactory for the prediction block of friendship and the overlapping than condition, obtain target prediction frame
The object detection method that the present embodiment proposes, by carrying out duplicate removal processing to the prediction block using preset algorithm,
To obtain target prediction frame；It realizes and duplicate removal processing is carried out to prediction block, to improve target detection rate.
Based on the 5th embodiment, the sixth embodiment of object detection method of the present invention is proposed, reference Fig. 8, in the present embodiment,
Step S43 includes：
Step S431 is obtained the confidence level of the prediction block, and is ranked up the prediction block based on the confidence level,
To obtain ranking results；
Step S432 is successively chosen based on the highest prediction block of previous belief and other prediction blocks by the ranking results
It calculates and hands over and compare；
Step S433 determines the friendship and than whether being greater than second threshold；
Step S434 eliminates the friendship and the prediction block than being greater than the second threshold, retains the friendship and ratio is not more than
The target prediction frame of second threshold.
In the present embodiment, when carrying out target detection using neural network, the corresponding confidence of each detection block will be exported
Degree, the confidence level of each detection block is ranked up, and calculates the corresponding prediction block of first confidence level using detection evaluation function
Friendship and ratio.
Further, it is determined that the friendship and than whether be greater than second threshold, if the friendship and than be greater than second threshold when, eliminate
Corresponding prediction block, if the friendship and than be less than or equal to second threshold when, retain the prediction block, for example, by confidence level from
Arriving small be ranked up greatly is 0.9,0.85,0.80, first calculating confidence level be the friendship of 0.9 corresponding prediction block and other prediction blocks simultaneously
Than determining the friendship and than then calculating the friendship of confidence level 0.85 corresponding prediction block and other prediction blocks whether greater than second threshold
And compare, it determines the friendship and is 0.80 corresponding prediction block and other prediction blocks than whether greater than second threshold, finally calculating confidence level
Friendship and ratio, determine the friendship and than whether be greater than second threshold, eliminate hand over and than be greater than the second preset threshold prediction block.
It is of course also possible to judge first confidence level, determine whether confidence level is less than a certain threshold value, retains confidence level
Less than the prediction block of the threshold value, then the corresponding confidence level of the prediction block is ranked up, it is pre- to calculate this further according to ranking results
It surveys the corresponding friendship of frame and compares, to eliminate the prediction block for being greater than second threshold.
Further, calculate the friendship of prediction block and than when, choose the maximum prediction block of confidence level as normative forecast frame,
The overlapping area and union area for calculating other prediction circles and normative forecast frame, determine the ratio of the overlapping area and union area
Whether value is greater than first threshold, if the ratio is greater than first threshold, eliminates the corresponding prediction circle of ratio in other prediction blocks,
If the ratio is not more than first threshold, retain the corresponding prediction block of ratio in other prediction circles, for example, prediction block includes square
Shape frame first assumes there are 6 rectangle frames, is sorted according to classifier category classification probability, be belonging respectively to the general of vehicle from small to large
Rate is respectively A, B, C, D, E, F, since maximum probability rectangle frame F, judges the friendship of A~E and F respectively and than whether being greater than some
The threshold value of setting, it is assumed that the degree of overlapping of B, D and F are more than threshold value, then just eliminating B, D；And first rectangle frame F of label, it carries out
Retain, then from remaining rectangle frame A, C, E, then the maximum E of select probability calculates the friendship of E and A, C and ratio, when handing over simultaneously
Than being greater than certain threshold value, then just eliminating；And marking E is second rectangle frame remained, and so on, it weighs always
It is multiple, find all rectangle frames being retained.
The object detection method that the present embodiment proposes by obtaining the confidence level of the prediction block, and is based on the confidence
The prediction block is ranked up by degree, to obtain ranking results, is then based on the ranking results and is successively chosen and work as previous belief
Highest prediction block calculates with other prediction blocks and hands over and compare, and then determines the friendship and than finally disappearing whether greater than second threshold
Except the friendship and the prediction block than being greater than the second threshold, retain the friendship and the target prediction than being not more than second threshold
Frame；It realizes and obtains final target detection frame using the method for non-maxima suppression and improve mesh so as to detect target
Target verification and measurement ratio.
The present invention also provides a kind of computer readable storage mediums, in the present embodiment, on computer readable storage medium
It is stored with target detection sequence, wherein：
Image to be detected is obtained, described image to be detected extracts feature by multilayer convolution in neural network and generates feature
Figure；
Modified structural parameters in neural network model are loaded, corresponding anchor frame is generated based on the structural parameters and sits
Mark, wherein the preset structure parameter includes the length-width ratio of the reference dimension of anchor frame, anchor frame scale and anchor frame；
Candidate frame coordinate is generated based on region nomination subnet, takes corresponding region to pass through according to candidate frame coordinate on characteristic pattern
It crosses area-of-interest pond and obtains individual features；
Prediction block coordinate is determined based on the feature, and target object location is determined based on the prediction block coordinate.
Further, when which is executed by the processor, following steps are also realized：
Area based on the reference dimension and the anchor frame dimension calculation anchor frame；
The length and width of anchor frame are calculated, based on the area and anchor frame length-width ratio to determine the size of anchor frame；
The centre coordinate of anchor frame is obtained, and determines the position of anchor frame based on the centre coordinate, generates anchor frame four edges
Coordinate.
Further, when which is executed by the processor, following steps are also realized：
Determine whether the anchor frame is more than original image range；
When the anchor frame is more than the original image range, the anchor frame more than original image range is eliminated, to obtain target anchor frame seat
Mark.
Further, when which is executed by the processor, following steps are also realized：
The feature obtains offset further across recurrence；
Final prediction block coordinate is obtained based on the corresponding target anchor frame coordinate of the offset correction, to determine object
Body position.
Further, when which is executed by the processor, following steps are also realized：
Duplicate removal processing is carried out to the prediction block using preset algorithm, to obtain target prediction frame.
Further, when which is executed by the processor, following steps are also realized：
The confidence level of the prediction block is obtained, and is ranked up the prediction block based on the confidence level, to be arranged
Sequence result；
It is successively chosen based on the ranking results and hands over and compare when the highest prediction block of previous belief is calculated with other prediction blocks；
Determine the friendship and than whether being greater than second threshold；
The friendship and the prediction block than being greater than the second threshold are eliminated, retains the friendship and than no more than second threshold
Target prediction frame.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the system that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or system institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or system.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in one as described above
In storage medium (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that terminal device (it can be mobile phone,
Computer, server, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (9)
1. a kind of object detection method, which is characterized in that the object detection method includes the following steps：
Image to be detected is obtained, described image to be detected extracts feature by multilayer convolution in neural network and generates characteristic pattern；
Modified structural parameters in neural network model are loaded, generate corresponding anchor frame coordinate based on the structural parameters,
In, the preset structure parameter includes the length-width ratio of the reference dimension of anchor frame, anchor frame scale and anchor frame；
Candidate frame coordinate is generated based on region nomination subnet, takes corresponding region by sense according to candidate frame coordinate on characteristic pattern
Interest pool area obtains individual features；
Prediction block coordinate is determined based on the feature, and target object location is determined based on the prediction block coordinate.
2. object detection method as described in claim 1, which is characterized in that described to generate anchor frame seat based on the structural parameters
Target step includes：
Area based on the reference dimension and the anchor frame dimension calculation anchor frame；
The length and width of anchor frame are calculated, based on the area and anchor frame length-width ratio to determine the size of anchor frame；
The centre coordinate of anchor frame is obtained, and determines the position of anchor frame based on the centre coordinate, generates the coordinate of anchor frame four edges.
3. object detection method as described in claim 1, which is characterized in that be based on the structural parameters and generate corresponding anchor
After the step of frame coordinate, the object detection method further includes：
Determine whether the anchor frame is more than original image range；
When the anchor frame is more than the original image range, the anchor frame more than original image range is eliminated, to obtain target anchor frame coordinate.
4. object detection method as claimed in claim 3, which is characterized in that described to determine that prediction block is sat based on the feature
Mark, and the step of determining target object location based on the prediction block coordinate includes：
The feature obtains offset further across recurrence；
Final prediction block coordinate is obtained based on the corresponding target anchor frame coordinate of the offset correction, to determine object position
It sets.
5. object detection method as claimed in claim 4, which is characterized in that described to be based on the corresponding mesh of the offset correction
Mark anchor frame coordinate obtains final prediction block coordinate, the step of to determine target object location after, the object detection method
Further include：
Duplicate removal processing is carried out to the prediction block using preset algorithm, to obtain target prediction frame.
6. object detection method as claimed in claim 5, which is characterized in that it is described using preset algorithm to the prediction block into
Row duplicate removal processing, to include the step of obtaining target prediction frame：
The confidence level of the prediction block is obtained, and is ranked up the prediction block based on the confidence level, to obtain sequence knot
Fruit；
It is successively chosen based on the ranking results and hands over and compare when the highest prediction block of previous belief is calculated with other prediction blocks；
Determine the friendship and than whether being greater than second threshold；
The friendship and the prediction block than being greater than the second threshold are eliminated, the friendship and the target than being not more than second threshold are retained
Prediction block.
7. object detection method as claimed in any one of claims 1 to 6, which is characterized in that the neural network includes object
Extractor, the candidate frame of body character pair generate network, candidate frame Region Feature Extraction, classification and Recurrent networks.
8. a kind of object detecting device, which is characterized in that the object detecting device includes：It memory, processor and is stored in
On the memory and the object detection program that can run on the processor, the object detection program is by the processor
The step of method as described in any one of claims 1 to 7 is realized when execution.
9. a kind of computer readable storage medium, which is characterized in that be stored with target inspection on the computer readable storage medium
Ranging sequence realizes the target detection as described in any one of claims 1 to 7 when the object detection program is executed by processor
Method and step.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810453480.5A CN108875577A (en) | 2018-05-11 | 2018-05-11 | Object detection method, device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810453480.5A CN108875577A (en) | 2018-05-11 | 2018-05-11 | Object detection method, device and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108875577A true CN108875577A (en) | 2018-11-23 |
Family
ID=64333590
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810453480.5A Pending CN108875577A (en) | 2018-05-11 | 2018-05-11 | Object detection method, device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108875577A (en) |
Cited By (54)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583512A (en) * | 2018-12-14 | 2019-04-05 | 北京旷视科技有限公司 | Image processing method, apparatus and system |
| CN109583396A (en) * | 2018-12-05 | 2019-04-05 | 广东亿迅科技有限公司 | A kind of region prevention method, system and terminal based on CNN two stages human testing |
| CN109710148A (en) * | 2018-12-19 | 2019-05-03 | 广州文远知行科技有限公司 | Image annotation frame selection method and device, computer equipment and storage medium |
| CN109886230A (en) * | 2019-02-28 | 2019-06-14 | 中南大学 | A kind of image object detection method and device |
| CN109886205A (en) * | 2019-02-25 | 2019-06-14 | 苏州清研微视电子科技有限公司 | Safety belt method of real-time and system |
| CN110009800A (en) * | 2019-03-14 | 2019-07-12 | 北京京东尚科信息技术有限公司 | A kind of recognition methods and equipment |
| CN110427940A (en) * | 2019-08-05 | 2019-11-08 | 山东浪潮人工智能研究院有限公司 | A method of pre-selection frame is generated for object detection model |
| CN110428357A (en) * | 2019-08-09 | 2019-11-08 | 厦门美图之家科技有限公司 | The detection method of watermark, device, electronic equipment and storage medium in image |
| CN110443212A (en) * | 2019-08-12 | 2019-11-12 | 睿魔智能科技（深圳）有限公司 | Positive sample acquisition methods, device, equipment and storage medium for target detection |
| CN110443366A (en) * | 2019-07-30 | 2019-11-12 | 上海商汤智能科技有限公司 | Optimization method and device, object detection method and the device of neural network |
| CN110738125A (en) * | 2019-09-19 | 2020-01-31 | 平安科技（深圳）有限公司 | Method, device and storage medium for selecting detection frame by using Mask R-CNN |
| CN110807385A (en) * | 2019-10-24 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN110910445A (en) * | 2019-11-26 | 2020-03-24 | 深圳市丰巢科技有限公司 | Object size detection method and device, detection equipment and storage medium |
| CN110930420A (en) * | 2019-11-11 | 2020-03-27 | 中科智云科技有限公司 | Dense target background noise suppression method and device based on neural network |
| CN111241875A (en) * | 2018-11-28 | 2020-06-05 | 驭势科技（北京）有限公司 | Automatic signboard semantic mapping and positioning method and system based on vision |
| CN111242119A (en) * | 2020-01-02 | 2020-06-05 | 腾讯科技（深圳）有限公司 | Vehicle image processing method and device, electronic equipment and computer readable medium |
| CN111291717A (en) * | 2020-02-28 | 2020-06-16 | 深圳前海微众银行股份有限公司 | Image-based object detection method, device, equipment and readable storage medium |
| CN111311673A (en) * | 2018-12-12 | 2020-06-19 | 北京京东尚科信息技术有限公司 | Positioning method and device and storage medium |
| CN111461182A (en) * | 2020-03-18 | 2020-07-28 | 北京小米松果电子有限公司 | Image processing method, image processing apparatus, and storage medium |
| CN111461068A (en) * | 2020-04-27 | 2020-07-28 | 湖南自兴智慧医疗科技有限公司 | Chromosome metaphase map identification and segmentation method |
| CN111476840A (en) * | 2020-05-14 | 2020-07-31 | 阿丘机器人科技(苏州)有限公司 | Target positioning method, device, equipment and computer readable storage medium |
| CN111488871A (en) * | 2019-01-25 | 2020-08-04 | 斯特拉德视觉公司 | Method and apparatus for switchable mode R-CNN based monitoring |
| CN111582257A (en) * | 2019-02-15 | 2020-08-25 | 波音公司 | Method, device and system for detecting object to be detected |
| CN111612726A (en) * | 2019-02-26 | 2020-09-01 | 广州慧睿思通信息科技有限公司 | Image data screening method and device, computer equipment and storage medium |
| CN111985267A (en) * | 2019-05-21 | 2020-11-24 | 长沙智能驾驶研究院有限公司 | Multi-target detection method, device, computer equipment, vehicle and storage medium |
| CN112052837A (en) * | 2020-10-09 | 2020-12-08 | 腾讯科技（深圳）有限公司 | Target detection method and device based on artificial intelligence |
| CN112101061A (en) * | 2019-06-17 | 2020-12-18 | 富士通株式会社 | Target detection method and device and image processing equipment |
| CN112215123A (en) * | 2020-10-09 | 2021-01-12 | 腾讯科技（深圳）有限公司 | Target detection method, device and storage medium |
| CN112308105A (en) * | 2019-08-02 | 2021-02-02 | 北京图森智途科技有限公司 | Target detection method, target detector and related equipment |
| CN112419263A (en) * | 2020-11-20 | 2021-02-26 | 上海电力大学 | Multi-class non-maximum inhibition method and system based on inter-class coverage ratio |
| CN112446374A (en) * | 2019-08-28 | 2021-03-05 | 北京地平线机器人技术研发有限公司 | Method and device for determining target detection model |
| CN112446231A (en) * | 2019-08-27 | 2021-03-05 | 丰图科技(深圳)有限公司 | Pedestrian crossing detection method and device, computer equipment and storage medium |
| CN112528907A (en) * | 2020-12-18 | 2021-03-19 | 四川云从天府人工智能科技有限公司 | Anchor frame generation and label frame adaptation method and device and computer storage medium |
| CN112528932A (en) * | 2020-12-22 | 2021-03-19 | 北京百度网讯科技有限公司 | Method and device for optimizing position information, road side equipment and cloud control platform |
| CN112560557A (en) * | 2019-09-25 | 2021-03-26 | 虹软科技股份有限公司 | People number detection method, face detection device and electronic equipment |
| CN112733652A (en) * | 2020-12-31 | 2021-04-30 | 深圳赛安特技术服务有限公司 | Image target identification method and device, computer equipment and readable storage medium |
| CN112749590A (en) * | 2019-10-30 | 2021-05-04 | 上海高德威智能交通系统有限公司 | Object detection method, device, computer equipment and computer readable storage medium |
| CN112749726A (en) * | 2020-02-26 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Training method and device of target detection model, computer equipment and storage medium |
| CN113076955A (en) * | 2021-04-14 | 2021-07-06 | 上海云从企业发展有限公司 | Target detection method, system, computer equipment and machine readable medium |
| CN113111709A (en) * | 2021-03-10 | 2021-07-13 | 北京爱笔科技有限公司 | Vehicle matching model generation method and device, computer equipment and storage medium |
| CN113190888A (en) * | 2021-04-20 | 2021-07-30 | 东风柳州汽车有限公司 | Process documentation method, device, equipment and storage medium |
| CN113326749A (en) * | 2021-05-17 | 2021-08-31 | 合肥高维数据技术有限公司 | Target detection method and device, storage medium and electronic equipment |
| CN113538407A (en) * | 2018-12-29 | 2021-10-22 | 北京市商汤科技开发有限公司 | Anchor point determining method and device, electronic equipment and storage medium |
| CN113632095A (en) * | 2019-03-16 | 2021-11-09 | 辉达公司 | Object detection using tilted polygons suitable for parking space detection |
| CN113781544A (en) * | 2020-07-14 | 2021-12-10 | 北京沃东天骏信息技术有限公司 | Plane detection method and device |
| CN114219930A (en) * | 2021-12-06 | 2022-03-22 | 安徽省配天机器人集团有限公司 | Feature point detection method, feature point detection device, and computer-readable storage medium |
| JP2022068146A (en) * | 2021-04-28 | 2022-05-09 | ベイジン バイドゥ ネットコム サイエンス テクノロジー カンパニー リミテッド | Method for annotating data, apparatus, storage medium, and computer program |
| WO2022148109A1 (en) * | 2021-01-05 | 2022-07-14 | 歌尔股份有限公司 | Product defect detection method and apparatus, device and computer-readable storage medium |
| CN114998705A (en) * | 2022-06-17 | 2022-09-02 | 集美大学 | Target detection method and system and in-memory computing chip |
| CN114998867A (en) * | 2022-06-27 | 2022-09-02 | 安徽酷哇机器人有限公司 | Traffic light state detection method and system based on single camera |
| CN115035186A (en) * | 2021-12-03 | 2022-09-09 | 荣耀终端有限公司 | Target object marking method and terminal equipment |
| CN115457036A (en) * | 2022-11-10 | 2022-12-09 | 中国平安财产保险股份有限公司 | Detection model training method, intelligent counting method and related equipment |
| CN115953605A (en) * | 2023-03-14 | 2023-04-11 | 深圳中集智能科技有限公司 | Machine vision multi-target image coordinate matching method |
| CN118115953A (en) * | 2024-04-29 | 2024-05-31 | 武汉大学中南医院 | Surgical hand disinfection intelligent monitoring method and device applicable to operating room |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316058A (en) * | 2017-06-15 | 2017-11-03 | 国家新闻出版广电总局广播科学研究院 | Improve the method for target detection performance by improving target classification and positional accuracy |
| CN107451602A (en) * | 2017-07-06 | 2017-12-08 | 浙江工业大学 | A kind of fruits and vegetables detection method based on deep learning |
| CN107610047A (en) * | 2017-08-02 | 2018-01-19 | 深圳市易成自动驾驶技术有限公司 | Fragmental image processing method, apparatus and computer-readable recording medium |
-
2018
- 2018-05-11 CN CN201810453480.5A patent/CN108875577A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316058A (en) * | 2017-06-15 | 2017-11-03 | 国家新闻出版广电总局广播科学研究院 | Improve the method for target detection performance by improving target classification and positional accuracy |
| CN107451602A (en) * | 2017-07-06 | 2017-12-08 | 浙江工业大学 | A kind of fruits and vegetables detection method based on deep learning |
| CN107610047A (en) * | 2017-08-02 | 2018-01-19 | 深圳市易成自动驾驶技术有限公司 | Fragmental image processing method, apparatus and computer-readable recording medium |
Non-Patent Citations (2)
| Title |
|---|
| ROSS GIRSHICK 等: "Fast R-CNN", 《ARXIV》 * |
| SHAOQING REN 等: "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks", 《ARXIV》 * |
Cited By (78)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111241875A (en) * | 2018-11-28 | 2020-06-05 | 驭势科技（北京）有限公司 | Automatic signboard semantic mapping and positioning method and system based on vision |
| CN109583396A (en) * | 2018-12-05 | 2019-04-05 | 广东亿迅科技有限公司 | A kind of region prevention method, system and terminal based on CNN two stages human testing |
| CN111311673B (en) * | 2018-12-12 | 2023-11-03 | 北京京东乾石科技有限公司 | Positioning method and device and storage medium |
| CN111311673A (en) * | 2018-12-12 | 2020-06-19 | 北京京东尚科信息技术有限公司 | Positioning method and device and storage medium |
| CN109583512A (en) * | 2018-12-14 | 2019-04-05 | 北京旷视科技有限公司 | Image processing method, apparatus and system |
| CN109583512B (en) * | 2018-12-14 | 2021-05-25 | 北京旷视科技有限公司 | Image processing method, device and system |
| CN109710148A (en) * | 2018-12-19 | 2019-05-03 | 广州文远知行科技有限公司 | Image annotation frame selection method and device, computer equipment and storage medium |
| CN113538407B (en) * | 2018-12-29 | 2022-10-14 | 北京市商汤科技开发有限公司 | Anchor point determining method and device, electronic equipment and storage medium |
| CN113538407A (en) * | 2018-12-29 | 2021-10-22 | 北京市商汤科技开发有限公司 | Anchor point determining method and device, electronic equipment and storage medium |
| CN111488871A (en) * | 2019-01-25 | 2020-08-04 | 斯特拉德视觉公司 | Method and apparatus for switchable mode R-CNN based monitoring |
| CN111488871B (en) * | 2019-01-25 | 2023-08-04 | 斯特拉德视觉公司 | Method and apparatus for R-CNN based monitoring of switchable modes |
| CN111582257A (en) * | 2019-02-15 | 2020-08-25 | 波音公司 | Method, device and system for detecting object to be detected |
| CN109886205B (en) * | 2019-02-25 | 2023-08-08 | 苏州清研微视电子科技有限公司 | Real-time safety belt monitoring method and system |
| CN109886205A (en) * | 2019-02-25 | 2019-06-14 | 苏州清研微视电子科技有限公司 | Safety belt method of real-time and system |
| CN111612726A (en) * | 2019-02-26 | 2020-09-01 | 广州慧睿思通信息科技有限公司 | Image data screening method and device, computer equipment and storage medium |
| CN109886230A (en) * | 2019-02-28 | 2019-06-14 | 中南大学 | A kind of image object detection method and device |
| CN110009800A (en) * | 2019-03-14 | 2019-07-12 | 北京京东尚科信息技术有限公司 | A kind of recognition methods and equipment |
| CN113632095A (en) * | 2019-03-16 | 2021-11-09 | 辉达公司 | Object detection using tilted polygons suitable for parking space detection |
| CN111985267A (en) * | 2019-05-21 | 2020-11-24 | 长沙智能驾驶研究院有限公司 | Multi-target detection method, device, computer equipment, vehicle and storage medium |
| CN111985267B (en) * | 2019-05-21 | 2024-09-27 | 长沙智能驾驶研究院有限公司 | Multi-target detection method, device, computer equipment, vehicle and storage medium |
| CN112101061A (en) * | 2019-06-17 | 2020-12-18 | 富士通株式会社 | Target detection method and device and image processing equipment |
| CN110443366A (en) * | 2019-07-30 | 2019-11-12 | 上海商汤智能科技有限公司 | Optimization method and device, object detection method and the device of neural network |
| CN112308105B (en) * | 2019-08-02 | 2024-04-12 | 北京图森智途科技有限公司 | Target detection method, target detector and related equipment |
| CN112308105A (en) * | 2019-08-02 | 2021-02-02 | 北京图森智途科技有限公司 | Target detection method, target detector and related equipment |
| CN110427940A (en) * | 2019-08-05 | 2019-11-08 | 山东浪潮人工智能研究院有限公司 | A method of pre-selection frame is generated for object detection model |
| CN110428357A (en) * | 2019-08-09 | 2019-11-08 | 厦门美图之家科技有限公司 | The detection method of watermark, device, electronic equipment and storage medium in image |
| CN110443212A (en) * | 2019-08-12 | 2019-11-12 | 睿魔智能科技（深圳）有限公司 | Positive sample acquisition methods, device, equipment and storage medium for target detection |
| CN110443212B (en) * | 2019-08-12 | 2022-03-11 | 睿魔智能科技（深圳）有限公司 | Positive sample acquisition method, device, equipment and storage medium for target detection |
| CN112446231A (en) * | 2019-08-27 | 2021-03-05 | 丰图科技(深圳)有限公司 | Pedestrian crossing detection method and device, computer equipment and storage medium |
| CN112446374A (en) * | 2019-08-28 | 2021-03-05 | 北京地平线机器人技术研发有限公司 | Method and device for determining target detection model |
| CN112446374B (en) * | 2019-08-28 | 2023-08-29 | 北京地平线机器人技术研发有限公司 | Method and device for determining target detection model |
| CN110738125B (en) * | 2019-09-19 | 2023-08-01 | 平安科技（深圳）有限公司 | Method, device and storage medium for selecting detection frame by Mask R-CNN |
| CN110738125A (en) * | 2019-09-19 | 2020-01-31 | 平安科技（深圳）有限公司 | Method, device and storage medium for selecting detection frame by using Mask R-CNN |
| CN112560557A (en) * | 2019-09-25 | 2021-03-26 | 虹软科技股份有限公司 | People number detection method, face detection device and electronic equipment |
| CN110807385B (en) * | 2019-10-24 | 2024-01-12 | 腾讯科技（深圳）有限公司 | Target detection method, target detection device, electronic equipment and storage medium |
| CN110807385A (en) * | 2019-10-24 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN112749590A (en) * | 2019-10-30 | 2021-05-04 | 上海高德威智能交通系统有限公司 | Object detection method, device, computer equipment and computer readable storage medium |
| CN112749590B (en) * | 2019-10-30 | 2023-02-07 | 上海高德威智能交通系统有限公司 | Object detection method, device, computer equipment and computer readable storage medium |
| CN110930420A (en) * | 2019-11-11 | 2020-03-27 | 中科智云科技有限公司 | Dense target background noise suppression method and device based on neural network |
| CN110930420B (en) * | 2019-11-11 | 2022-09-30 | 中科智云科技有限公司 | Dense target background noise suppression method and device based on neural network |
| CN110910445A (en) * | 2019-11-26 | 2020-03-24 | 深圳市丰巢科技有限公司 | Object size detection method and device, detection equipment and storage medium |
| CN111242119A (en) * | 2020-01-02 | 2020-06-05 | 腾讯科技（深圳）有限公司 | Vehicle image processing method and device, electronic equipment and computer readable medium |
| CN111242119B (en) * | 2020-01-02 | 2022-12-16 | 腾讯科技（深圳）有限公司 | Vehicle image processing method and device, electronic equipment and computer readable medium |
| CN112749726A (en) * | 2020-02-26 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Training method and device of target detection model, computer equipment and storage medium |
| CN112749726B (en) * | 2020-02-26 | 2023-09-29 | 腾讯科技（深圳）有限公司 | Training method and device for target detection model, computer equipment and storage medium |
| CN111291717A (en) * | 2020-02-28 | 2020-06-16 | 深圳前海微众银行股份有限公司 | Image-based object detection method, device, equipment and readable storage medium |
| CN111461182A (en) * | 2020-03-18 | 2020-07-28 | 北京小米松果电子有限公司 | Image processing method, image processing apparatus, and storage medium |
| CN111461182B (en) * | 2020-03-18 | 2023-04-18 | 北京小米松果电子有限公司 | Image processing method, image processing apparatus, and storage medium |
| CN111461068A (en) * | 2020-04-27 | 2020-07-28 | 湖南自兴智慧医疗科技有限公司 | Chromosome metaphase map identification and segmentation method |
| CN111461068B (en) * | 2020-04-27 | 2023-10-17 | 湖南自兴智慧医疗科技有限公司 | Chromosome metaphase map identification and segmentation method |
| CN111476840A (en) * | 2020-05-14 | 2020-07-31 | 阿丘机器人科技(苏州)有限公司 | Target positioning method, device, equipment and computer readable storage medium |
| CN111476840B (en) * | 2020-05-14 | 2023-08-22 | 阿丘机器人科技(苏州)有限公司 | Target positioning method, device, equipment and computer readable storage medium |
| CN113781544B (en) * | 2020-07-14 | 2024-05-24 | 北京沃东天骏信息技术有限公司 | Plane detection method and device |
| CN113781544A (en) * | 2020-07-14 | 2021-12-10 | 北京沃东天骏信息技术有限公司 | Plane detection method and device |
| CN112052837A (en) * | 2020-10-09 | 2020-12-08 | 腾讯科技（深圳）有限公司 | Target detection method and device based on artificial intelligence |
| CN112215123A (en) * | 2020-10-09 | 2021-01-12 | 腾讯科技（深圳）有限公司 | Target detection method, device and storage medium |
| CN112419263A (en) * | 2020-11-20 | 2021-02-26 | 上海电力大学 | Multi-class non-maximum inhibition method and system based on inter-class coverage ratio |
| CN112528907B (en) * | 2020-12-18 | 2024-04-09 | 四川云从天府人工智能科技有限公司 | Anchor frame generation and label frame adaptation method and device and computer storage medium |
| CN112528907A (en) * | 2020-12-18 | 2021-03-19 | 四川云从天府人工智能科技有限公司 | Anchor frame generation and label frame adaptation method and device and computer storage medium |
| CN112528932B (en) * | 2020-12-22 | 2023-12-08 | 阿波罗智联(北京)科技有限公司 | Method and device for optimizing position information, road side equipment and cloud control platform |
| CN112528932A (en) * | 2020-12-22 | 2021-03-19 | 北京百度网讯科技有限公司 | Method and device for optimizing position information, road side equipment and cloud control platform |
| CN112733652A (en) * | 2020-12-31 | 2021-04-30 | 深圳赛安特技术服务有限公司 | Image target identification method and device, computer equipment and readable storage medium |
| CN112733652B (en) * | 2020-12-31 | 2024-04-19 | 深圳赛安特技术服务有限公司 | Image target recognition method, device, computer equipment and readable storage medium |
| WO2022148109A1 (en) * | 2021-01-05 | 2022-07-14 | 歌尔股份有限公司 | Product defect detection method and apparatus, device and computer-readable storage medium |
| CN113111709B (en) * | 2021-03-10 | 2023-12-29 | 北京爱笔科技有限公司 | Vehicle matching model generation method, device, computer equipment and storage medium |
| CN113111709A (en) * | 2021-03-10 | 2021-07-13 | 北京爱笔科技有限公司 | Vehicle matching model generation method and device, computer equipment and storage medium |
| CN113076955A (en) * | 2021-04-14 | 2021-07-06 | 上海云从企业发展有限公司 | Target detection method, system, computer equipment and machine readable medium |
| CN113190888A (en) * | 2021-04-20 | 2021-07-30 | 东风柳州汽车有限公司 | Process documentation method, device, equipment and storage medium |
| JP2022068146A (en) * | 2021-04-28 | 2022-05-09 | ベイジン バイドゥ ネットコム サイエンス テクノロジー カンパニー リミテッド | Method for annotating data, apparatus, storage medium, and computer program |
| CN113326749A (en) * | 2021-05-17 | 2021-08-31 | 合肥高维数据技术有限公司 | Target detection method and device, storage medium and electronic equipment |
| CN115035186A (en) * | 2021-12-03 | 2022-09-09 | 荣耀终端有限公司 | Target object marking method and terminal equipment |
| CN114219930A (en) * | 2021-12-06 | 2022-03-22 | 安徽省配天机器人集团有限公司 | Feature point detection method, feature point detection device, and computer-readable storage medium |
| CN114998705A (en) * | 2022-06-17 | 2022-09-02 | 集美大学 | Target detection method and system and in-memory computing chip |
| CN114998705B (en) * | 2022-06-17 | 2024-09-24 | 集美大学 | Target detection method, system and in-memory computing chip |
| CN114998867A (en) * | 2022-06-27 | 2022-09-02 | 安徽酷哇机器人有限公司 | Traffic light state detection method and system based on single camera |
| CN115457036A (en) * | 2022-11-10 | 2022-12-09 | 中国平安财产保险股份有限公司 | Detection model training method, intelligent counting method and related equipment |
| CN115953605A (en) * | 2023-03-14 | 2023-04-11 | 深圳中集智能科技有限公司 | Machine vision multi-target image coordinate matching method |
| CN118115953A (en) * | 2024-04-29 | 2024-05-31 | 武汉大学中南医院 | Surgical hand disinfection intelligent monitoring method and device applicable to operating room |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108875577A (en) | Object detection method, device and computer readable storage medium | |
| CN108596101B (en) | Remote sensing image multi-target detection method based on convolutional neural network | |
| CN108898047B (en) | Pedestrian detection method and system based on blocking and shielding perception | |
| CN109815770B (en) | Two-dimensional code detection method, device and system | |
| CN107330437B (en) | Feature extraction method based on convolutional neural network target real-time detection model | |
| CN109670452A (en) | Method for detecting human face, device, electronic equipment and Face datection model | |
| CN109816745B (en) | Human body thermodynamic diagram display method and related products | |
| CN108229332A (en) | Bone attitude determination method, device and computer readable storage medium | |
| CN109147254A (en) | A kind of video outdoor fire disaster smog real-time detection method based on convolutional neural networks | |
| CN109948590A (en) | Pose problem detection method and device | |
| CN108052624A (en) | Processing Method of Point-clouds, device and computer readable storage medium | |
| CN109410275A (en) | Position predicting method, device, equipment and readable storage medium storing program for executing | |
| CN103514432A (en) | Method, device and computer program product for extracting facial features | |
| CN107818288A (en) | Sign board information acquisition method and device | |
| CN106778502A (en) | A kind of people counting method based on depth residual error network | |
| CN107679503A (en) | A kind of crowd's counting algorithm based on deep learning | |
| CN109815843A (en) | Object detection method and Related product | |
| CN108960229A (en) | One kind is towards multidirectional character detecting method and device | |
| CN111986250A (en) | Object volume measuring method, device, measuring equipment and storage medium | |
| CN104463240B (en) | A kind of instrument localization method and device | |
| CN116152863B (en) | Personnel information identification method and device, electronic equipment and storage medium | |
| CN107464290A (en) | Three-dimensional information methods of exhibiting, device and mobile terminal | |
| CN110910445B (en) | Object size detection method, device, detection equipment and storage medium | |
| CN106874913A (en) | A kind of vegetable detection method | |
| CN113763440A (en) | Image processing method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181123 |
|
| RJ01 | Rejection of invention patent application after publication |