CN108021926A - A kind of vehicle scratch detection method and system based on panoramic looking-around system - Google Patents
A kind of vehicle scratch detection method and system based on panoramic looking-around system Download PDFInfo
- Publication number
- CN108021926A CN108021926A CN201710904966.1A CN201710904966A CN108021926A CN 108021926 A CN108021926 A CN 108021926A CN 201710904966 A CN201710904966 A CN 201710904966A CN 108021926 A CN108021926 A CN 108021926A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- deep learning
- network model
- training sample
- learning network
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- 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
- G06N3/08—Learning methods
Abstract
A kind of vehicle scratch detection method and system based on panoramic looking-around system, the deep learning neural network model of Accurate classification can be carried out by there is label training sample set pair deep learning network model to be trained acquisition, the vehicle body image gathered using the model to the panoramic looking-around system through distortion correction and pretreatment is handled, and obtains the scratch situation of vehicle.The present invention can automatic identification vehicle scratch, judge driver safety driving condition, provide differentiated products for automobile leasing and car insurance industry and offer foundation is provided.By to there is the selection of label training sample, increasing the data enhancing processing to training sample image and the training to deep learning network model, the present invention has the recognition accuracy of higher compared to conventional images identification technology.It coordinates panoramic looking-around system, can effectively avoid the appearance that scratch situation is omitted in artificial detection.
Description
Technical field
The present invention relates to image identification technical field, more particularly to a kind of vehicle to scrape detection method and system.
Background technology
In the fast development of national communication, vehicle guaranteeding organic quantity increases sharply.Vehicle scratch in automobile leasing and
It is to judge that driver uses one of important references index of vehicle safety situation in car insurance industry.Based on vehicle scratch feelings
Condition, can effectively estimate drive vehicle lose or upkeep cost so that instruct automobile leasing or insurance provider provided for client
The product of differentiation or service.
But at this stage, the correlative study of vehicle scratch detection is less, and the scratch of daily vehicle still relies on work substantially
Personnel visually observe what detection determined.But visually observe inevitably can because scratch is slight or vision interference and caused by detect dredge
Suddenly.Moreover, existing artificial detection needs to expend a large amount of costs of labor, detection efficiency is low.
With the fast development of image and computer vision technique, image recognition technology is applied to automobile more and more
Electronic field.Vehicle-mounted camera can not only provide the environmental information of vehicle periphery, be also used as video, image data acquiring
End, data are provided for post analysis vehicle body status.But the conventional images identification technology indication character trickle to vehicle scratch etc. is difficult
To be accurately distinguished, it is not enough to be directly used in the detection of vehicle scratch.
Therefore, at present, it is badly in need of a kind of method or system that can monitor vehicle scratch automatically, improves the high score of vehicle scratch
Class precision, accurate judgement vehicle scratch situation.
The content of the invention
In order to solve the shortcomings of the prior art, it is an object of the invention to provide a kind of based on panoramic looking-around system
Vehicle scratch detection method and system.
First, to achieve the above object, propose a kind of vehicle scratch detection method, comprise the following steps:
The first step, gathers vehicle body image;
Second step, carries out distortion correction and pretreatment to the vehicle body image, obtains data set;The pretreatment includes:
To the data normalization processing of the vehicle body image after completion distortion correction and whitening pretreatment;
3rd step, inputs trained deep learning network model in advance by second step the data obtained collection, carries out feature and carry
Take and classify;
4th step, output category result, obtains vehicle scratch situation.
Further, in the 3rd step of above-mentioned vehicle scratch detection method, the deep learning network model is in advance by following
Training step obtains:
S1, establishment have label training sample set { (X1,y1), (X2,y2) ..., (Xk,yk), wherein, k is specimen number, k
>=2, XkRepresent k-th of training sample image, ykRepresent the label value of k-th of training sample;It is described to have label training sample set bag
At least one positive sample and at least one negative sample are included, the label value of the positive sample is different from the label value of the negative sample;Institute
The scale for stating each training sample image is normalized to identical size;
S2, training sample image pretreatment, including data normalization processing and albefaction to each training sample image
Pretreatment;
S3, establishes deep learning network model, sets the initiation parameter of the deep learning network model;
S4, successively inputs pretreated each training sample image in the step S2 to the deep learning network
Model is trained, and feedforward calculating is carried out to the deep learning network model of initiation parameter, according to result of calculation and label value
ykDifference obtain error in classification, using gradient descent algorithm carry out error back propagation computing, adjust the deep learning net
Each parameter value of network model;Replace training sample and repeat above procedure progress new round training, until the error in classification is less than
Threshold value set in advance, obtains trained deep learning network model;
S5, when each parameter of trained deep learning network model in the step S4 is deployed into the measurement scratch inspection
In survey method.
Further, in the step S2 of above-mentioned vehicle scratch detection method, further include to each training sample image
Data enhancing is handled;The data enhancing processing includes:The scaling of the training sample image, rotation, inclination, contrast are become
Change, or more one or more data enhancing processing combination.
Specifically, in above-mentioned vehicle scratch detection method, the whitening pretreatment includes ZCA whitening pretreatments, PCA albefactions
Pretreatment or other whitening pretreatment technologies.
Preferably to train the deep learning network model, in above-mentioned vehicle scratch detection method, the negative sample bag
Include but be not limited to include the training sample image that leaf, spot etc. are easy to obscure with vehicle scratch.
Further, in the step S3 of above-mentioned vehicle scratch detection method, the deep learning network model is using convolution god
Through network struction.
Secondly, to achieve the above object, it is also proposed that a kind of vehicle scratch detecting system, includes the flake shooting of interconnection
Head, storage device and server;
The fish-eye camera is used to gather vehicle body image and store to the storage device or be uploaded to the server;
The storage device is used to store the vehicle body image;
The server is used for:First, the vehicle body image is read, distortion correction and pre- place are carried out to the vehicle body image
Reason, obtains data set;Wherein, the pretreatment includes:To complete distortion correction after vehicle body view data normalized and
Whitening pretreatment;Then, by the advance trained deep learning network model of the data obtained collection input after pretreatment, feature is carried out
Extraction and classification;Finally, classification results are calculated, export vehicle scratch situation.
Specifically, in above-mentioned vehicle scratch detecting system, the fish-eye camera quantity is at least 2, and the flake is taken the photograph
Include the complete vehicle body side to the vehicle as the visual field of head.
Wherein, described two fish-eye cameras are respectively arranged at the bottom of the left and right rearview mirror of the vehicle, the flake
The field of view angle of camera is at least 180 °.
Further, in above-mentioned vehicle scratch detecting system, the server is onboard servers or remote server;It is described
Remote server is read and is identified the vehicle body image stored in the storage device by wireless network.
Beneficial effect
The present invention first detects deep learning network application in vehicle scratch, the vehicle body image by server to collection
Automatic decision is carried out, whether detection vehicle body has scratch.The present invention can overcome the inherent shortcoming of hand inspection, using institute of the present invention
The image detecting method based on deep learning network stated, can obtain the discrimination accuracy than traditional images detection method higher.
This method eligible result can be carried as the foundation judged driver safety driving condition for automobile leasing and car insurance industry
Scientific basis is provided for differentiated products and service.
Further, the present invention by the training to confusing negative sample, can effectively distinguish leaf, dirt in the training stage
The trace that stain etc. is easily obscured with scratch.Strengthen processing step by increasing data, increase training sample data amount, suppressed plan
Close.By whitening pretreatment to input data dimensionality reduction, data redundancy is reduced, further suppresses over-fitting.
Vehicle scratch detecting system provided by the present invention, can directly acquire whole vehicle body side by panoramic looking-around system
Complete image, effectively prevent the dead angle in scratch detection, improve the confidence level of result.Meanwhile the panoramic looking-around system
In, the lower distal end that two language cameras are arranged at vehicle mirrors is sentenced and just photographs complete vehicle body image.
Further, it is contemplated that the computing capability of onboard servers in itself, wireless network also may be selected in the present invention, by panorama ring
The vehicle body image that viewing system collects is uploaded to remote server, and scratch detection is carried out by remote server.So may be used
The installation cost of vehicle is saved, can also be saved to the renewal of deep learning network model and lower deployment cost in server.
Other features and advantages of the present invention will be illustrated in the following description, also, partly becomes from specification
Obtain it is clear that or being understood by implementing the present invention.
Brief description of the drawings
Attached drawing is used for providing a further understanding of the present invention, and a part for constitution instruction, and with the present invention's
Embodiment together, for explaining the present invention, is not construed as limiting the invention.In the accompanying drawings:
Fig. 1 is the flow chart according to vehicle scratch detection method of the present invention;
Fig. 2 is to be illustrated according to the training flow of deep learning network model in vehicle scratch detection method of the present invention
Figure;
Fig. 3 is the block diagram according to vehicle scratch detecting system of the present invention;
Fig. 4 is the schematic diagram according to a kind of deep learning network model of the embodiment of the present invention.
Embodiment
The preferred embodiment of the present invention is illustrated below in conjunction with attached drawing, it will be appreciated that described herein preferred real
Apply example to be merely to illustrate and explain the present invention, be not intended to limit the present invention.
Fig. 1 is according to the vehicle scratch detection method of the present invention, is comprised the following steps:
The first step, gathers vehicle body image;
Second step, carries out distortion correction and pretreatment to the vehicle body image, obtains data set;The pretreatment includes:
To the data normalization processing of the vehicle body image after completion distortion correction and whitening pretreatment;Here pre-treatment step is mainly
By to input data dimensionality reduction, reducing data redundancy, suppressing over-fitting;
3rd step, by second step the data obtained collection input in advance trained deep learning network model, carry out feature extraction and
Classification;
4th step, output category result, obtains vehicle scratch situation.
Vehicle body image distortion correction step in the second step, using spherical projection model method, such method calculates
The smaller and precision of amount is higher, and detailed process is as follows:
Spherical coordinate model foundation conventional coordinates in panorama picture of fisheye lens principle, and adjust its position and side
To so that fish-eye camera is located at coordinate origin O, and along Oz axis positive directions, the fish eye images after shooting fall shooting direction
In Oxy planes.
It is then determined that the coordinate conversion relation formula between correction chart picture and fish eye images:
Wherein, (x, y, z) is original image 3D coordinate points, and image coordinate point after (u, v) correction, R is panorama picture of fisheye lens original
The spherical radius of spherical coordinate model in reason.
Each pixel of correction chart picture is traveled through successively, and corresponding fish eye images pixel is obtained according to coordinate transformation relation
Point, is assigned to the pixel on correction chart picture, it is possible to obtain correction chart picture by pixel value.
Further, with reference to 2 flow of figure, in the 3rd step of above-mentioned vehicle scratch detection method, the deep learning network mould
Type is obtained by following training step in advance:
S1, establishment have label training sample set { (X1,y1), (X2,y2) ..., (Xk,yk), wherein, k is specimen number, k
>=2, XkRepresent k-th of training sample image, ykRepresent the label value of k-th of training sample;It is described to have label training sample set bag
At least one positive sample and at least one negative sample are included, the label value of the positive sample is different from the label value of the negative sample;Institute
Stating the scale of each training sample image, to normalize pixel value be 224*224;It is all from view of the image for being actually needed detection
Identical fish-eye camera, the pixel quantity that image is included have been fixed as same scale, have been pertained only in detection process to quiet
The identification of state image is without regard to processing video, therefore without dimension normalization processing, but while training needs to carry out on scale
Normalization.
S2, training sample image pretreatment, including data normalization processing and albefaction to each training sample image
Pretreatment;Here, the data normalization processing of image refers specifically to sample-by-sample average abatement:For three color channels of coloured image
In each passage, respectively calculate average pixel value of the passage in whole image region, then with each passage
Specific each pixel value subtracts the average pixel value of this passage.The image caused by the difference of illumination condition can so be eliminated
Between difference.
S3, establishes deep learning network model, sets the initiation parameter of the deep learning network model;
S4, successively inputs pretreated each training sample image in the step S2 to the deep learning network
Model is trained, and feedforward calculating is carried out to the deep learning network model of initiation parameter, according to result of calculation and label value
ykDifference obtain error in classification, using gradient descent algorithm carry out error back propagation computing, adjust the deep learning net
Each parameter value of network model, replaces training sample and repeats above procedure progress new round training, until the error in classification is less than
Threshold value set in advance, obtains trained deep learning network model;
S5, when each parameter of trained deep learning network model in the step S4 is deployed into the measurement scratch inspection
In survey method.
Further, it is contemplated that the image data amount of scratch is smaller, and it is more smart that training samples number is typically not enough to support
True identification requirement is, it is necessary to suppress over-fitting, therefore, in the step S2 of above-mentioned vehicle scratch detection method, usually also
(if training sample is sufficient, can be omitted this step) is handled including the data enhancing to each training sample image;Institute
Stating data enhancing processing includes:To the scaling of the training sample image, rotation, inclination, contrast variation, or more it is a kind of or
The combination of a variety of data enhancing processing.Random zoomed image ± 10% specifically may be selected, rotated at random in the range of 0 to 360 degree
Image, it is random to tilt
Further, in the second step of above-mentioned vehicle scratch detection method or the step S2, the whitening pretreatment bag
Include ZCA whitening pretreatments, PCA whitening pretreatments or other whitening pretreatment technologies.
One of which PCA whitening pretreatment methods, it is comprised the following steps that:
(1) average value of image pattern pixel value is calculated
(2) Eigen Covariance matrix is calculated
(3) Eigen Covariance characteristic value and feature vector are solved
(4) k maximum characteristic value is chosen
(5) sample point is projected in the feature vector of selection
Preferably to train the deep learning network model, in above-mentioned vehicle scratch detection method, the negative sample bag
Include but be not limited to include the training sample image that leaf, spot etc. are easy to obscure with vehicle scratch.
Wherein, in reference Fig. 4, the step S3 of above-mentioned vehicle scratch detection method, the deep learning network model uses
Convolutional neural networks are built, and specifically on the basis of the LeNet-5 network architectures, last layer is changed to one-dimensional vector, passes through
The classification output of sigmoid functions two, the network include 7 layers altogether.
Secondly, the connection relation with reference to shown in Fig. 3, a kind of vehicle scratch is also proposed based on the above method, in the present embodiment
Detecting system, including the fish-eye camera of interconnection, storage device and server;
The fish-eye camera is used to gather vehicle body image and store to the storage device or be uploaded to the server;
The storage device is used to store the vehicle body image;
The server is used for:First, the vehicle body image is read, distortion correction and pre- place are carried out to the vehicle body image
Reason, obtains data set;Wherein, the pretreatment includes:To complete distortion correction after vehicle body image dimension normalization processing,
Data normalization processing and whitening pretreatment;Then, by the advance trained deep learning of the data obtained collection input after pretreatment
Network model, carries out feature extraction and classification;Finally, classification results are calculated, export vehicle scratch situation.
Specifically, in above-mentioned vehicle scratch detecting system, the fish-eye camera quantity is at least 2, can use 4 fishes
Eye imaging head forms panoramic looking-around system, and the visual field of the panoramic looking-around system includes the complete vehicle body side to the vehicle.
Wherein, described two fish-eye cameras are respectively arranged at the bottom of the left and right rearview mirror of the vehicle, the flake
The field of view angle of camera is at least 180 °.
Further, in above-mentioned vehicle scratch detecting system, the server is onboard servers or remote server;It is described
Remote server is read and is identified the vehicle body image stored in the storage device by wireless network.
The advantages of technical solution of the present invention, is mainly reflected in:The present invention is first by deep learning network application in vehicle scratch
Detection, carries out the vehicle body image of collection automatic decision, whether detection vehicle body has scratch by server.The present invention can overcome
The inherent shortcoming of hand inspection, using the image detecting method of the present invention based on deep learning network, can obtain than passing
The discrimination accuracy for image detecting method higher of uniting.By the training to confusing negative sample, this method can effectively distinguish tree
The trace that leaf, spot etc. are easily obscured with scratch.This method eligible result can as driver safety driving condition is judged according to
According to for automobile leasing and car insurance industry provides differentiated products and service provides scientific basis.
One of ordinary skill in the art will appreciate that:The foregoing is only a preferred embodiment of the present invention, and does not have to
In the limitation present invention, although the present invention is described in detail with reference to the foregoing embodiments, for those skilled in the art
For, its still can to foregoing embodiments record technical solution modify, or to which part technical characteristic into
Row equivalent substitution.Within the spirit and principles of the invention, any modification, equivalent replacement, improvement and so on, should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of vehicle scratch detection method, it is characterised in that step includes:
The first step, gathers vehicle body image;
Second step, carries out distortion correction and pretreatment to the vehicle body image, obtains data set;The pretreatment includes:To complete
Data normalization processing and whitening pretreatment into the vehicle body image after distortion correction;
3rd step, by second step the data obtained collection input in advance trained deep learning network model, carry out feature extraction and
Classification;
4th step, output category result, obtains vehicle scratch situation.
2. vehicle scratch detection method as claimed in claim 1, it is characterised in that in the 3rd step, the deep learning network
Model is obtained by following training step in advance:
S1, establishment have label training sample set { (X1,y1), (X2,y2) ..., (Xk,yk), wherein, k is specimen number, k >=2, Xk
Represent k-th of training sample image, ykRepresent the label value of k-th of training sample;It is described to there is label training sample set to include extremely
Few 1 positive sample and at least one negative sample, the label value of the positive sample are different from the label value of the negative sample;It is described each
The scale of training sample image is normalized to identical size;
S2, training sample image pre-process, including the data normalization processing and albefaction to each training sample image are located in advance
Reason;
S3, establishes deep learning network model, sets the initiation parameter of the deep learning network model;
S4, successively inputs pretreated each training sample image in the step S2 to the deep learning network model
It is trained, feedforward calculating is carried out to the deep learning network model of initiation parameter, according to result of calculation and label value yk's
Difference obtains error in classification, carries out error back propagation computing using gradient descent algorithm, adjusts the deep learning network mould
Each parameter value of type;Replace training sample and repeat above procedure progress new round training, until the error in classification is less than advance
The threshold value of setting, obtains trained deep learning network model;
S5, when each parameter of trained deep learning network model in the step S4 is deployed into the measurement scratch detection side
In method.
3. vehicle scratch detection method as claimed in claim 2, it is characterised in that in step S2, further include to each instruction
Practice the data enhancing processing of sample image;The data enhancing processing includes:To the scaling of the training sample image, rotation,
Inclination, contrast variation, or more one or more data enhancing processing combination.
4. vehicle scratch detection method as claimed in claim 2, it is characterised in that in the second step or the step S2,
The whitening pretreatment includes ZCA whitening pretreatments or PCA whitening pretreatments.
5. vehicle scratch detection method as claimed in claim 2, it is characterised in that the negative sample includes but not limited to include
The training sample image of leaf or spot.
6. vehicle scratch detection method as claimed in claim 2, it is characterised in that in the step S3, the deep learning
Network model is built using convolutional neural networks.
7. a kind of vehicle scratch detecting system, it is characterised in that fish-eye camera, storage device and service including interconnection
Device;
The fish-eye camera is used to gather vehicle body image and store to the storage device or be uploaded to the server;
The storage device is used to store the vehicle body image
The server is used for:First, the vehicle body image is read, distortion correction and pretreatment are carried out to the vehicle body image,
Obtain data set;Wherein, the pretreatment includes:The data normalization of vehicle body image after completion distortion correction is handled and white
Change pretreatment;Then, by the data obtained collection input after pretreatment, trained deep learning network model, progress feature carry in advance
Take and classify;Finally, classification results are calculated, export vehicle scratch situation.
8. vehicle scratch detecting system as claimed in claim 7, it is characterised in that the fish-eye camera quantity is at least 2
A, the visual field of the fish-eye camera includes the complete vehicle body side to the vehicle.
9. vehicle scratch detecting system as claimed in claim 8, it is characterised in that described two fish-eye cameras are set respectively
In the bottom of the left and right rearview mirror of the vehicle, the field of view angle of the fish-eye camera is at least 180 °.
10. the vehicle scratch detecting system as described in claim 7 to 9, it is characterised in that the server is onboard servers
Or remote server;The remote server is read and is identified the vehicle body figure stored in the storage device by wireless network
Picture.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710904966.1A CN108021926A (en) | 2017-09-28 | 2017-09-28 | A kind of vehicle scratch detection method and system based on panoramic looking-around system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710904966.1A CN108021926A (en) | 2017-09-28 | 2017-09-28 | A kind of vehicle scratch detection method and system based on panoramic looking-around system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108021926A true CN108021926A (en) | 2018-05-11 |
Family
ID=62079540
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710904966.1A Pending CN108021926A (en) | 2017-09-28 | 2017-09-28 | A kind of vehicle scratch detection method and system based on panoramic looking-around system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108021926A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108956653A (en) * | 2018-05-31 | 2018-12-07 | 广东正业科技股份有限公司 | A kind of quality of welding spot detection method, system, device and readable storage medium storing program for executing |
CN108986249A (en) * | 2018-06-26 | 2018-12-11 | 杭州车厘子智能科技有限公司 | Vehicle remote damage identification method and system based on panoramic looking-around image |
CN109086716A (en) * | 2018-08-01 | 2018-12-25 | 北京嘀嘀无限科技发展有限公司 | A kind of method and device of seatbelt wearing detection |
CN110308091A (en) * | 2019-07-11 | 2019-10-08 | 中体彩印务技术有限公司 | A kind of detection method and system of anti-counterfeiting mark |
CN110348381A (en) * | 2019-07-11 | 2019-10-18 | 电子科技大学 | A kind of video behavior recognition methods based on deep learning |
CN110717401A (en) * | 2019-09-12 | 2020-01-21 | Oppo广东移动通信有限公司 | Age estimation method and device, equipment and storage medium |
CN112560964A (en) * | 2020-12-18 | 2021-03-26 | 深圳赛安特技术服务有限公司 | Method and system for training Chinese herbal medicine pest and disease identification model based on semi-supervised learning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102854191A (en) * | 2012-07-18 | 2013-01-02 | 湖南大学 | Real-time visual detection and identification method for high speed rail surface defect |
KR20130074860A (en) * | 2011-12-27 | 2013-07-05 | 건국대학교 산학협력단 | Device and method for restorating degraded regions in movie film |
CN104992449A (en) * | 2015-08-06 | 2015-10-21 | 西安冉科信息技术有限公司 | Information identification and surface defect on-line detection method based on machine visual sense |
CN105279556A (en) * | 2015-11-05 | 2016-01-27 | 国家卫星海洋应用中心 | Enteromorpha detection method and enteromorpha detection device |
CN106156765A (en) * | 2016-08-30 | 2016-11-23 | 南京邮电大学 | safety detection method based on computer vision |
CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A kind of metal plate and belt detection method of surface flaw and device based on Faster R CNN networks |
-
2017
- 2017-09-28 CN CN201710904966.1A patent/CN108021926A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20130074860A (en) * | 2011-12-27 | 2013-07-05 | 건국대학교 산학협력단 | Device and method for restorating degraded regions in movie film |
CN102854191A (en) * | 2012-07-18 | 2013-01-02 | 湖南大学 | Real-time visual detection and identification method for high speed rail surface defect |
CN104992449A (en) * | 2015-08-06 | 2015-10-21 | 西安冉科信息技术有限公司 | Information identification and surface defect on-line detection method based on machine visual sense |
CN105279556A (en) * | 2015-11-05 | 2016-01-27 | 国家卫星海洋应用中心 | Enteromorpha detection method and enteromorpha detection device |
CN106156765A (en) * | 2016-08-30 | 2016-11-23 | 南京邮电大学 | safety detection method based on computer vision |
CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A kind of metal plate and belt detection method of surface flaw and device based on Faster R CNN networks |
Non-Patent Citations (2)
Title |
---|
FAN W等: "An automatic machine vision method for the flaw detection on car's body", 《IEEE INTERNATIONAL CONFERENCE ON AWARENESS SCIENCE AND TECHNOLOGY》 * |
王保丰,刘传凯: "《月球车遥操作中的计算机视觉技术》", 31 May 2016, 国防工业出版社 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108956653A (en) * | 2018-05-31 | 2018-12-07 | 广东正业科技股份有限公司 | A kind of quality of welding spot detection method, system, device and readable storage medium storing program for executing |
CN108986249A (en) * | 2018-06-26 | 2018-12-11 | 杭州车厘子智能科技有限公司 | Vehicle remote damage identification method and system based on panoramic looking-around image |
CN108986249B (en) * | 2018-06-26 | 2021-08-03 | 杭州车厘子智能科技有限公司 | Vehicle remote damage assessment method and system based on panoramic all-around image |
CN109086716A (en) * | 2018-08-01 | 2018-12-25 | 北京嘀嘀无限科技发展有限公司 | A kind of method and device of seatbelt wearing detection |
CN110308091A (en) * | 2019-07-11 | 2019-10-08 | 中体彩印务技术有限公司 | A kind of detection method and system of anti-counterfeiting mark |
CN110348381A (en) * | 2019-07-11 | 2019-10-18 | 电子科技大学 | A kind of video behavior recognition methods based on deep learning |
CN110717401A (en) * | 2019-09-12 | 2020-01-21 | Oppo广东移动通信有限公司 | Age estimation method and device, equipment and storage medium |
CN112560964A (en) * | 2020-12-18 | 2021-03-26 | 深圳赛安特技术服务有限公司 | Method and system for training Chinese herbal medicine pest and disease identification model based on semi-supervised learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108021926A (en) | A kind of vehicle scratch detection method and system based on panoramic looking-around system | |
CN109308693B (en) | Single-binocular vision system for target detection and pose measurement constructed by one PTZ camera | |
CN108171673B (en) | Image processing method and device, vehicle-mounted head-up display system and vehicle | |
US8005264B2 (en) | Method of automatically detecting and tracking successive frames in a region of interesting by an electronic imaging device | |
JP6305171B2 (en) | How to detect objects in a scene | |
CN111144207B (en) | Human body detection and tracking method based on multi-mode information perception | |
CN109407547A (en) | Multi-cam assemblage on-orbit test method and system towards panoramic vision perception | |
CN111476827B (en) | Target tracking method, system, electronic device and storage medium | |
CN110889829B (en) | Monocular distance measurement method based on fish eye lens | |
CN109145864A (en) | Determine method, apparatus, storage medium and the terminal device of visibility region | |
JP2011198349A (en) | Method and apparatus for processing information | |
CN110232389A (en) | A kind of stereoscopic vision air navigation aid based on green crop feature extraction invariance | |
CN110414381A (en) | Tracing type face identification system | |
CN111965636A (en) | Night target detection method based on millimeter wave radar and vision fusion | |
CN110543848B (en) | Driver action recognition method and device based on three-dimensional convolutional neural network | |
CN114022560A (en) | Calibration method and related device and equipment | |
CN111797684A (en) | Binocular vision distance measuring method for moving vehicle | |
CN114170537A (en) | Multi-mode three-dimensional visual attention prediction method and application thereof | |
CN110956065B (en) | Face image processing method and device for model training | |
CN116894775B (en) | Bolt image preprocessing method based on camera motion model recovery and super-resolution | |
CN112070077A (en) | Deep learning-based food identification method and device | |
CN113723432B (en) | Intelligent identification and positioning tracking method and system based on deep learning | |
Li et al. | Vision-based target detection and positioning approach for underwater robots | |
CN109978779A (en) | A kind of multiple target tracking device based on coring correlation filtering method | |
CN115601538A (en) | Target detection method, target detector, target detection system, and 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: 20180511 |
|
RJ01 | Rejection of invention patent application after publication |