CN110705553A - Scratch detection method suitable for vehicle distant view image - Google Patents

Scratch detection method suitable for vehicle distant view image Download PDF

Info

Publication number
CN110705553A
CN110705553A CN201911013854.2A CN201911013854A CN110705553A CN 110705553 A CN110705553 A CN 110705553A CN 201911013854 A CN201911013854 A CN 201911013854A CN 110705553 A CN110705553 A CN 110705553A
Authority
CN
China
Prior art keywords
scratch
area
region
color
training
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.)
Granted
Application number
CN201911013854.2A
Other languages
Chinese (zh)
Other versions
CN110705553B (en
Inventor
王新年
王淏
齐国清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian Maritime University
Original Assignee
Dalian Maritime University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Dalian Maritime University filed Critical Dalian Maritime University
Priority to CN201911013854.2A priority Critical patent/CN110705553B/en
Publication of CN110705553A publication Critical patent/CN110705553A/en
Application granted granted Critical
Publication of CN110705553B publication Critical patent/CN110705553B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • 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
    • 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
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • 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
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Health & Medical Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Signal Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a scratch detection method suitable for a vehicle long-range image, which comprises the following steps: segmenting an interested area possibly having scratches by using a deep learning algorithm; obtaining candidate scratch regions in the region of interest by utilizing a scratch detection and MSER method fusing color and spatial information; and screening the candidate scratch area by comprehensively utilizing a Hough line detection and SVM method, and further marking the scratch area. The invention comprehensively uses a plurality of image processing methods to detect the scratch area in the region of interest, and avoids the problems that the scratch area in the vehicle distant view image is too small to cause unobvious characteristics and the scratch and the interference area are difficult to distinguish. Meanwhile, the invention does not need to shoot close-range images manually, and relieves manual operation from complex scratch detection tasks.

Description

Scratch detection method suitable for vehicle distant view image
Technical Field
The invention relates to the field of image processing and deep learning, in particular to a scratch detection method suitable for a vehicle long-range image.
Background
At present, research specially aiming at the detection of the scratch on the surface of the vehicle is lacked, and the research aiming at the detection of the damage of the vehicle is similar to the lacked research. The existing vehicle damage detection methods can be mainly divided into two types, namely a damage detection method based on computer vision and a damage detection method based on deep learning, wherein the vehicle damage detection method based on deep learning can be divided into a detection method based on a residual error dense network and a detection method based on a Faster R-CNN target identification algorithm. The main ideas of the methods are as follows:
(1) the vehicle damage judging method based on computer vision comprises the following steps: firstly, arranging a binocular image acquisition system, and calibrating by using a calibration plate; then, collecting the vehicle image in the monitoring area and obtaining a depth map of the collected image; then, taking the depth map as a training set, training a convolutional neural network and obtaining a vehicle damage degree discrimination model; and finally, judging the damage degree of the newly acquired vehicle image by using a judgment model.
(2) The vehicle damage detection method based on the residual error dense network comprises the following steps: firstly, processing an input image for multiple times through a dense residual error network to obtain a global feature map of the input image; and then detecting the global characteristic diagram through a damage detection model based on a single-point multi-box recognizer algorithm, and marking the detected vehicle damage.
(3) The vehicle damage detection method based on the Faster R-CNN target identification algorithm comprises the following steps: firstly, marking a damaged area in a training image as a foreground, and marking other areas as backgrounds; inputting the marked training image into a Faster R-CNN neural network for training to obtain a damage model; and finally, inputting a test image, detecting a damage area in the test image by using the damage model and marking the damage area.
The computer vision-based vehicle scratch detection algorithm has the problems that: two cameras with the same model are required to be fixed on an optical platform at a certain baseline distance for building a binocular image acquisition system, so that an observation target is ensured to be within the imaging range of the two cameras, and calibration is carried out by utilizing a calibration plate. The cost of building a system is high, the preparation work is tedious, and the system is not suitable for large-scale application.
The vehicle scratch detection algorithm based on deep learning has the following problems: the detection accuracy is high when the close view image of the vehicle is processed, but the detection accuracy is greatly reduced for the far view image of the vehicle. However, close-up images need to be shot manually, i.e. the existing method must be added with manual operation to be used. When the deep learning-based vehicle scratch detection algorithm is used for processing a long-range image of a vehicle, the detection accuracy rate can be greatly reduced because: the interference in the image increases when processing the distant view image, such as a pattern on the clothes of pedestrians, a gap between bricks on the ground, a gap at the connection of parts on the vehicle body, and the like. In addition to increased disturbance, the feature is not apparent due to too small a scratch area. In the long-range image of the vehicle, the scratch occupies only a small part of the image, and for example, a 1080 × 960 long-range image, the scratch occupies only 180 pixels on average in the image, i.e., occupies only about 0.02% of the image area. In order to reduce the data amount of the feature map in the neural network, the size of an image is greatly reduced after the image is subjected to convolution, and in other words, in the case of downsampling, a scratch which is originally small in area is greatly reduced in area after the image is downsampled, and may even disappear from the feature map. This results in the scratch being less distinctive in character and difficult to distinguish from the disturbed area.
Disclosure of Invention
According to the technical problems of high use cost and poor detection precision in the prior art, the scratch detection method suitable for the vehicle long-range image is provided. The long-range image refers to: the single image at least comprises an image of the whole automobile, and a long-range view image and a short-range view image of the automobile are respectively shown in fig. 3 and fig. 4. The vehicle scratch refers to: the vehicle paint is damaged without damaging the vehicle damage caused by the sheet metal parts. The invention comprehensively uses a plurality of image processing methods to detect the scratch area in the region of interest, thereby avoiding the problems that the scratch area in the vehicle distant view image is too small to cause unobvious characteristics and the scratch and the interference area are difficult to distinguish.
The technical means adopted by the invention are as follows:
a scratch detection method suitable for a vehicle perspective image is characterized by comprising the following steps:
s1, detecting the possibly appearing interested region of the scratch by using a deep learning algorithm;
s2, respectively carrying out scratch detection on the region of interest by utilizing a scratch detection model fusing color and spatial information and an MSER method to obtain two candidate region sets, and marking as R1And R2Taking the intersection R of the two sets as a new candidate region set, namely R ═ R1∩R2；
And S3, screening the candidate region set R by comprehensively utilizing a Hough line detection and SVM method, and further marking a scratch region.
Further, the method also comprises a scratch detection model for training an extraction model of the region of interest and fusing color and spatial information.
Further, the training of the extraction model of the region of interest includes:
inputting training images, and scaling all the training images into a uniform size;
marking out an interested area in the training image, wherein the interested area is an area where scratches possibly appear on the vehicle body;
and inputting the marked training image into a Mask R-CNN neural network for training to obtain an extraction model of the region of interest.
Further, training the scratch detection model fusing color and spatial information includes:
in the training process, counting color information of scratches in all training images to obtain a Gaussian distribution model of the colors of the scratches, and meanwhile, counting color mean value information in neighborhoods of each point of the scratches to obtain a Gaussian distribution model of the color mean value in the neighborhoods of the scratches;
in the testing process, comprehensively considering the color information of the current point and the color mean value information of the pixel points in the neighborhood, respectively calculating the probability densities of the current point and the pixel points in the Gaussian distribution model, and judging that the color of the current point is matched with the color of the scratch to be a true scratch when the geometric mean value of the two probability densities is greater than a set threshold value, or judging the current point to be a false scratch;
and repeating the process for all the pixel points in the region of interest to complete the scratch screening of the fused color and space information.
Further, the screening the candidate scratch region by comprehensively utilizing a Hough line detection and SVM method comprises the following steps:
carrying out Hough line detection on the new candidate scratch area, and deleting a linear interference area with the detected length exceeding a set threshold value from the candidate scratch area;
and constructing reserved candidate scratch areas as connected areas, sequentially calculating two-dimensional characteristic values of each connected area, performing secondary classification on the two-dimensional characteristics by using a trained classification hyperplane, judging whether the area belongs to the scratch, and marking if the area belongs to the scratch to finish the vehicle scratch detection process.
Further, the step of obtaining the classification hyperplane comprises:
marking out real and false scratch areas in the training image;
calculating two-dimensional characteristic values of the true scratch area and the false scratch area, wherein the two-dimensional characteristic values comprise circularity and area filling degree;
and training a true scratch classifier and a false scratch classifier based on the SVM by using the calculated two-dimensional characteristic value to obtain a classification hyperplane.
Compared with the prior art, the invention has the following advantages:
1) according to the invention, a binocular image acquisition system is not required to be built, and the scratch detection task can be completed only by a common camera, so that the labor cost for building a hardware system can be reduced.
2) According to the method, the interested region in the image is segmented by using deep learning, so that the interference of the surrounding environment on scratch detection can be greatly reduced, and the problem of reduction of the detection accuracy caused by excessive interference in the vehicle long-range image is solved.
3) The invention comprehensively uses a plurality of image processing methods to detect the scratch area in the region of interest, and the image processing methods have the advantage of insensitivity to the area of the processing area, thereby avoiding the problems that the scratch area is too small in the vehicle distant view image, the characteristics are not obvious, and the scratch and the interference area are difficult to distinguish.
4) The invention is suitable for long-range images of vehicles, can place a camera in a parking lot to collect images, and then completes a full-automatic scratch detection task by utilizing the invention. The invention does not need to shoot close-range images manually, and relieves manual operation from complex scratch detection tasks.
For the reasons, the invention can be widely popularized in the fields of shared automobile management, parking lot management and the like.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of the model training process of the present invention.
FIG. 2 is a flow chart of the testing process of the present invention.
Fig. 3 is a perspective view of a scratch on a vehicle according to an embodiment of the present invention.
Fig. 4 is a close-up view of a scratch mark on a vehicle according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention discloses a scratch detection method suitable for a vehicle long-range image, which comprises the following steps:
s1, segmenting an interested area possibly having scratches by using a deep learning algorithm;
s2, respectively aligning the scratch detection model and the MSER method by using the fusion color and the spatial informationScratch detection is carried out on the region of interest to obtain two candidate region sets which are marked as R1And R2The intersection of the two sets is used as a new candidate region set and is denoted as R, that is, R is R1∩R2；
And S3, screening the candidate region set R by comprehensively utilizing a Hough line detection and SVM method, and further marking a scratch region. The method comprises the following steps: carrying out Hough line detection on the new candidate scratch area, and deleting a linear interference area with the detected length exceeding a set threshold value from the candidate scratch area; and constructing reserved candidate scratch areas as connected areas, sequentially calculating two-dimensional characteristic values of each connected area, performing secondary classification on the two-dimensional characteristics by using a trained classification hyperplane, judging whether the area belongs to the scratch, and marking if the area belongs to the scratch to finish the vehicle scratch detection process. The step of obtaining the classification hyperplane comprises the following steps: marking out real and false scratch areas in the training image; calculating two-dimensional characteristic values of the true scratch area and the false scratch area, wherein the two-dimensional characteristic values comprise circularity and area filling degree; and training a true scratch classifier and a false scratch classifier based on the SVM by using the calculated two-dimensional characteristic value to obtain a classification hyperplane.
Further, the method also comprises the step of training an extraction model of the region of interest and a scratch detection model fusing color and spatial information. Wherein the training of the extraction model of the region of interest comprises: inputting training images, and scaling all the training images into a uniform size; marking out an interested area in the training image, wherein the interested area is an area where scratches possibly appear on the vehicle body; and inputting the marked training image into a Mask R-CNN neural network for training to obtain an extraction model of the region of interest. Training the scratch detection model fusing color and spatial information comprises: in the training process, counting color information of scratches in all training images to obtain a Gaussian distribution model of the colors of the scratches, and meanwhile, counting color mean value information in neighborhoods of each point of the scratches to obtain a Gaussian distribution model of the color mean value in the neighborhoods of the scratches; in the testing process, comprehensively considering the color information of the current point and the color mean value information of the pixel points in the neighborhood, respectively calculating the probability densities of the current point and the pixel points in the Gaussian distribution model, and judging that the color of the current point is matched with the color of the scratch to be a true scratch when the geometric mean value of the two probability densities is greater than a set threshold value, or judging the current point to be a false scratch; and repeating the process for all the pixel points in the region of interest to complete the scratch screening of the fused color and space information.
According to the method, a Mask R-CNN deep learning algorithm is utilized to segment out regions of interest, namely regions where scratches are likely to appear, such as vehicle doors, front bumpers and the like; then, segmenting candidate scratch Regions in the region of interest by utilizing scratch detection and MSER (maximum Stable extreme region extraction) which are fused with color and space information; then, screening the candidate scratch area by comprehensively utilizing methods such as Hough line detection, SVM (Support Vector Machine) and the like; finally, marking a scratch area.
Further, before the region of interest where the scratch may appear is segmented by using the deep learning algorithm, the method further comprises the step of carrying out equalization processing on the test image to remove the influence of the illumination intensity on the brightness and the color of the image.
The invention is described in detail below with reference to the figures and examples.
As shown in fig. 1 and fig. 2, a scratch detection method suitable for a perspective image of a vehicle mainly includes two processes:
1) training process: inputting the marked training image into Mask R-CNN for training to obtain an extraction model of the region of interest; inputting the circularity and the area filling degree of the true scratch and the false scratch into an SVM (support vector machine), and obtaining a classification hyperplane for distinguishing the true scratch and the false scratch; and counting the scratch color information and the neighborhood color mean value information thereof to further obtain a Gaussian distribution model of the scratch color information and the neighborhood color mean value information. The specific process is as follows:
① input training images, all of which are scaled uniformly to 1280 x 960.
②, the regions of interest in the training images are marked, which are areas of the vehicle body where scratches may appear, such as doors, front and rear bumpers, while other areas of the vehicle body, such as wheels, headlights, windshields, etc., are not considered.
③, inputting the marked training image into a Mask R-CNN neural network for training to obtain a training model of the region of interest, wherein the Mask R-CNN is a general example segmentation architecture, which not only can correctly find the target in the image, but also can precisely segment the target, namely, a Mask covering the target precisely is generated, and other segmentation architectures only can generate a circumscribed rectangle.
④ mark areas of true and false scratches in the training image, which are objects that are similar in appearance to the true scratches and can cause significant disturbances in the detection, such as patterns on pedestrian clothing, gaps between floor bricks, gaps at the junctions of various parts of the vehicle body, etc.
⑤, counting the color of the scratch in all the training images, calculating to obtain the mean value mu and covariance matrix sigma of the scratch color, forming a Gaussian distribution model of the scratch color, and recording as g (x; mu, sigma)nSum covariance matrix ∑nForming a Gaussian distribution model of the scratch neighborhood color mean value, and recording the model as g (x; mu)n,∑n)。
⑥ two-dimensional feature values, i.e., circularity and region fill, are calculated for all regions identified in procedure ④, as follows:
wherein, FcRepresents the circularity, L represents the perimeter of the connected region, and a represents the area of the connected region.
Figure BDA0002245048170000072
Wherein, FpdRepresents the region filling degree, A represents the area of the connected region, ArRepresenting the area of the circumcircle of the connected region.
⑦ train the SVM based true and false scratch classifiers using the calculated circularity and area filling of the true and false scratch areas.
2) The testing process comprises the following steps: inputting a test image, segmenting an interested area in the test image, detecting a candidate scratch area in the interested area, screening the candidate scratch area, and marking a real scratch, wherein the specific process comprises the following steps:
① inputting test image, and performing image equalization process to the test image to remove the influence of light intensity on image brightness and color.
② the region of interest in the test image is segmented using the region of interest extraction model obtained by the training process.
③ the scratch detection process of fusing color and spatial information is as follows:
first, in the region of interest, the corresponding probability density of the color of the current point in the Gaussian distribution model g (x; mu, sigma) is calculated and is marked as p (x)0,y0) The g (x; mu, sigma) refers to the gaussian distribution model of the scratch color obtained from process ⑤ during the training process.
Then, the color mean value of the pixel points in the neighborhood of the current point (preferably 8 neighborhoods) is calculated, and the mean value in a Gaussian distribution model g (x; mu) is obtainedn,∑n) The corresponding probability density in (1), is denoted as pn(x0,y0). G (x; mu) as describedn,∑n) The Gaussian distribution model of the scratch neighborhood color mean value obtained in the process ⑤ in the training process
Finally, p (x) is calculated0,y0) And pn(x0,y0) The calculation method is as follows:
Figure BDA0002245048170000081
when pi (x)0,y0) And when the color of the current point is larger than the set threshold value, judging that the color of the current point is matched with the color of the scratch.
And repeating the process for all the pixel points in the region of interest to complete the scratch detection of the fused color and space information. Detected regionDomain composition candidate region set R1。
④ Gray processing the equalized image to obtain gray image, extracting MSER maximum stable extremum region to divide regions with gray value variation rule similar to scratch2。
⑤R1、R2And taking intersection of the two candidate region sets to obtain a new candidate region set, and recording the new candidate region set as R, namely R is R1∩R2；。
⑥, the Hough line detection is carried out on candidate scratch areas, and the area which has the length exceeding the set threshold (for example, 300 pixels) and is in the shape of a straight line is regarded as a gap at the connection part of the components on the vehicle body, the candidate scratch areas meeting the above conditions are judged as interference areas, and the areas are deleted from the candidate scratch areas.
⑦, constructing the reserved candidate scratch area as a connected area, sequentially calculating two-dimensional characteristic values of each connected area, namely circularity and area filling degree, performing secondary classification on the two-dimensional characteristics by using a classification hyperplane trained by an SVM classifier, and judging whether the area belongs to the scratch or not.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (6)

1. A scratch detection method suitable for a vehicle perspective image is characterized by comprising the following steps:
s1, detecting the possibly appearing interested region of the scratch by using a deep learning algorithm;
s2, respectively carrying out scratch detection on the region of interest by utilizing a scratch detection model fusing color and spatial information and an MSER method to obtain two candidate region sets, and marking as R1And R2Taking the intersection R of the two sets as a new candidate region set, namely R ═ R1∩R2；
And S3, screening the candidate region set R by comprehensively utilizing a Hough line detection and SVM method, and further marking a scratch region.
2. The scratch detection method according to claim 1, further comprising training an extraction model of the region of interest and a scratch detection model that fuses color and spatial information.
3. The scratch detection method according to claim 2, wherein the training of the extraction model of the region of interest comprises:
inputting training images, and scaling all the training images into a uniform size;
marking out an interested area in the training image, wherein the interested area is an area where scratches possibly appear on the vehicle body;
and inputting the marked training image into a Mask R-CNN neural network for training to obtain an extraction model of the region of interest.
4. The scratch detection method according to claim 1 or 2, wherein the training of the scratch detection model fusing color and spatial information comprises:
in the training process, counting color information of scratches in all training images to obtain a Gaussian distribution model of the colors of the scratches, and meanwhile, counting color mean value information in neighborhoods of each point of the scratches to obtain a Gaussian distribution model of the color mean value in the neighborhoods of the scratches;
in the testing process, comprehensively considering the color information of the current point and the color mean value information of the pixel points in the neighborhood, respectively calculating the probability densities of the current point and the pixel points in the Gaussian distribution model, and judging that the color of the current point is matched with the color of the scratch to be a true scratch when the geometric mean value of the two probability densities is greater than a set threshold value, or judging the current point to be a false scratch;
and repeating the process for all the pixel points in the region of interest to complete the scratch screening of the fused color and space information.
5. The scratch detection method according to claim 1, wherein the screening the candidate scratch regions by using a Hough line detection and SVM method in combination comprises:
carrying out Hough line detection on the new candidate scratch area, and deleting a linear interference area with the detected length exceeding a set threshold value from the candidate scratch area;
and constructing reserved candidate scratch areas as connected areas, sequentially calculating two-dimensional characteristic values of each connected area, performing secondary classification on the two-dimensional characteristics by using a trained classification hyperplane, judging whether the area belongs to the scratch, and marking if the area belongs to the scratch to finish the vehicle scratch detection process.
6. The scratch detection method according to claim 5, wherein the step of obtaining the classification hyperplane comprises:
marking out real and false scratch areas in the training image;
calculating two-dimensional characteristic values of the true scratch area and the false scratch area, wherein the two-dimensional characteristic values comprise circularity and area filling degree;
and training a true scratch classifier and a false scratch classifier based on the SVM by using the calculated two-dimensional characteristic value to obtain a classification hyperplane.
CN201911013854.2A 2019-10-23 2019-10-23 Scratch detection method suitable for vehicle distant view image Active CN110705553B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911013854.2A CN110705553B (en) 2019-10-23 2019-10-23 Scratch detection method suitable for vehicle distant view image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911013854.2A CN110705553B (en) 2019-10-23 2019-10-23 Scratch detection method suitable for vehicle distant view image

Publications (2)

Publication Number Publication Date
CN110705553A true CN110705553A (en) 2020-01-17
CN110705553B CN110705553B (en) 2022-10-25

Family

ID=69201447

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911013854.2A Active CN110705553B (en) 2019-10-23 2019-10-23 Scratch detection method suitable for vehicle distant view image

Country Status (1)

Country Link
CN (1) CN110705553B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112001294A (en) * 2020-08-19 2020-11-27 福建船政交通职业学院 YOLACT + + based vehicle body surface damage detection and mask generation method and storage device
CN112785561A (en) * 2021-01-07 2021-05-11 天津狮拓信息技术有限公司 Second-hand commercial vehicle condition detection method based on improved Faster RCNN prediction model
CN112861952A (en) * 2021-01-29 2021-05-28 云南电网有限责任公司电力科学研究院 Partial discharge image matching deep learning method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106295636A (en) * 2016-07-21 2017-01-04 重庆大学 Passageway for fire apparatus based on multiple features fusion cascade classifier vehicle checking method
US20180165541A1 (en) * 2016-12-12 2018-06-14 Jack Cooper Logistics, LLC System, method, and apparatus for detection of damages on surfaces

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106295636A (en) * 2016-07-21 2017-01-04 重庆大学 Passageway for fire apparatus based on multiple features fusion cascade classifier vehicle checking method
US20180165541A1 (en) * 2016-12-12 2018-06-14 Jack Cooper Logistics, LLC System, method, and apparatus for detection of damages on surfaces

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李克斌等: "基于形态学特征的机械零件表面划痕检测", 《光学学报》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112001294A (en) * 2020-08-19 2020-11-27 福建船政交通职业学院 YOLACT + + based vehicle body surface damage detection and mask generation method and storage device
CN112001294B (en) * 2020-08-19 2023-05-12 福建船政交通职业学院 Vehicle body surface damage detection and mask generation method and storage device based on YOLACT++
CN112785561A (en) * 2021-01-07 2021-05-11 天津狮拓信息技术有限公司 Second-hand commercial vehicle condition detection method based on improved Faster RCNN prediction model
CN112861952A (en) * 2021-01-29 2021-05-28 云南电网有限责任公司电力科学研究院 Partial discharge image matching deep learning method
CN112861952B (en) * 2021-01-29 2023-04-28 云南电网有限责任公司电力科学研究院 Partial discharge image matching deep learning method

Also Published As

Publication number Publication date
CN110705553B (en) 2022-10-25

Similar Documents

Publication Publication Date Title
CN108171112B (en) Vehicle identification and tracking method based on convolutional neural network
CN109101924B (en) Machine learning-based road traffic sign identification method
CN106709436B (en) Track traffic panoramic monitoring-oriented cross-camera suspicious pedestrian target tracking system
CN107545239B (en) Fake plate detection method based on license plate recognition and vehicle characteristic matching
CN105488454B (en) Front vehicles detection and ranging based on monocular vision
CN102509098B (en) Fisheye image vehicle identification method
CN106683119B (en) Moving vehicle detection method based on aerial video image
CN110705553B (en) Scratch detection method suitable for vehicle distant view image
CN110544211B (en) Method, system, terminal and storage medium for detecting lens attached object
CN108256521B (en) Effective area positioning method for vehicle body color identification
CN107665327B (en) Lane line detection method and device
CN112488046B (en) Lane line extraction method based on high-resolution images of unmanned aerial vehicle
CN113393426B (en) Steel rolling plate surface defect detection method
CN107808524B (en) Road intersection vehicle detection method based on unmanned aerial vehicle
CN112734761B (en) Industrial product image boundary contour extraction method
CN108509950B (en) Railway contact net support number plate detection and identification method based on probability feature weighted fusion
Farag Real-time detection of road lane-lines for autonomous driving
CN113034378B (en) Method for distinguishing electric automobile from fuel automobile
CN109544535B (en) Peeping camera detection method and system based on optical filtering characteristics of infrared cut-off filter
Chen et al. Front vehicle detection and distance estimation using single-lens video camera
CN110348307B (en) Path edge identification method and system for crane metal structure climbing robot
CN111160231A (en) Automatic driving environment road extraction method based on Mask R-CNN
CN109146863A (en) A kind of pavement marker line defect detection device
CN116152758A (en) Intelligent real-time accident detection and vehicle tracking method
CN111339824A (en) Road surface sprinkled object detection method based on machine vision

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
GR01 Patent grant
GR01 Patent grant