Automobile appearance detection system

By using different vehicle appearance models and mobile detection equipment in the automotive appearance detection system, combined with convolutional neural network and driving mechanism, the detection of vehicle bottom damage is achieved, solving the problem that the existing system cannot detect vehicle bottom and improving detection coverage and accuracy.

CN118961707AInactive Publication Date: 2024-11-15SHENYANG INST OF ENG

Patent Information

Application Number
CN202410914419.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing automotive appearance inspection system cannot detect the bottom of the vehicle, which poses safety risks and has low coverage of the detection range.

Method used

Using different vehicle model appearance models and mobile detection equipment, the damaged or paint-disappearing areas are identified through image acquisition and convolutional neural networks, and comprehensive inspection is carried out through extended drives and surround drives, including coverage detection of the bottom of the vehicle.

Benefits of technology

It improves the meticulousness and accuracy of the inspection, increases the coverage of vehicle bottom damage, avoids safety hazards, and achieves comprehensive inspection of vehicle damage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118961707A_ABST
    Figure CN118961707A_ABST
Patent Text Reader

Abstract

The invention discloses an automobile appearance detection system, and belongs to the technical field of automobile appearance detection systems. Comprising the steps of collecting appearance data of different vehicle types, collecting data of different vehicle types and detection positions of a vehicle, performing preprocessing such as denoising on the appearance data of the different vehicle types so as to facilitate subsequent model processing, establishing appearance models of the different vehicle types by utilizing the appearance data of the different vehicle types, and performing training through the appearance data of the different vehicle types. The performance and generalization ability of the model are optimized by using the verification set of the appearance data of different vehicle types, the trained appearance models of different vehicle types are deployed in an actual application scene, images are acquired through mobile camera shooting, and the images are input into the appearance models of different vehicle types for detection of damaged or paint-removed areas. Through the arrangement of the appearance models of different vehicle types, the damage condition of the vehicle can be detected, and meanwhile, the damage area can be estimated, so that the detection meticulousness and accuracy are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of automobile appearance detection systems, and more specifically, to an automobile appearance detection system. Background Art

[0002] With the progress of society, cars have become popular in thousands of households and have become people's primary means of transportation. However, during driving, bumps and collisions are inevitable, which require inspection and repair.

[0003] The document with the prior art publication number CN115684172A provides a vehicle appearance detection system and a method for using the same. The device obtains image information through an image acquisition module, and then analyzes the image information through an image preprocessing module and an image processing module, thereby analyzing the defects of the vehicle appearance. The device includes: a detection bracket, an image acquisition module, an image storage module, an image preprocessing module, an image processing module and a controller; the detection bracket includes a base, a mobile frame, and a mobile component. The mobile frame is slidably arranged on the base, and the vehicle is located below the mobile frame. There are multiple mobile components, and the multiple mobile components are slidably arranged on the mobile frame. The mobile component is provided with an image acquisition module, the image acquisition module includes a camera and a light source, both of which are arranged on the mobile component. The camera is used to capture an image of the vehicle appearance under the control of the controller, and the light source provides illumination for the camera under the control of the controller. The image acquisition module also includes a distance sensor, which is arranged on the mobile component and is electrically connected to the camera. When the distance sensor detects that the distance between the vehicle and the mobile component is less than a preset value, the distance sensor sends a signal to the camera, and the camera takes a picture of the vehicle.

[0004] Although the above-mentioned prior art solution can achieve the relevant beneficial effects through the structure of the prior art, it still has the following defects: when the device is in use, it cannot detect the bottom of the vehicle, and if there is a bump on the bottom of the vehicle, there will be certain safety hazards, and the coverage of the detection range is low.

[0005] In view of this, we propose a car appearance inspection system. Summary of the invention

[0006] 1. Technical issues to be solved

[0007] The purpose of this application is to provide a car appearance inspection system, which solves the technical problem that the device in the above-mentioned background technology cannot detect the bottom of the vehicle when in use, and if there is a bump on the bottom of the vehicle, there will be certain safety hazards, and the coverage of the detection range is low, thereby achieving a technical effect.

[0008] 2. Technical solution

[0009] The technical solution of the present application provides a vehicle appearance detection system, including

[0010] Collect appearance data of different vehicle models, and collect data of different vehicle models and detection locations;

[0011] Preprocessing of appearance data of different models, such as denoising, for the convenience of subsequent model processing;

[0012] Establish appearance model training for different models, and use appearance data of different models to establish appearance models of different models and train them through appearance data of different models;

[0013] Optimize and tune the appearance models of different car models. Use the validation sets of appearance data of different car models to evaluate the performance of the trained appearance models of different car models on new data. Based on the evaluation results, adjust the hyperparameters to optimize the performance and generalization ability of the model.

[0014] Deployment of appearance models of different vehicle models: Deploy the trained appearance models of different vehicle models to actual application scenarios, acquire images through mobile cameras, and input the images into the appearance models of different vehicle models to detect damaged or peeling areas;

[0015] Mobile detection equipment will provide reminders of abnormal locations based on prediction results, and change locations for comprehensive testing.

[0016] By adopting the above technical solution and setting the appearance models of different vehicle models, the damage of the vehicle can be detected and the area of ​​the damage can be estimated, thereby improving the meticulousness and accuracy of the detection.

[0017] As an optional solution of the technical solution of this application document, the data of appearance of different models is collected by using a detection head;

[0018] The appearance image data of different car models are labeled, including intactness, degree of damage and model. At the same time, the damaged or paint-chipped areas in the image are marked.

[0019] As an optional solution of the technical solution of the present application document, the preprocessing of the appearance data of different models is to convert the appearance data image of different models into a binary image when the appearance data image of different models is not a binary image, that is, the damaged area has been marked as white, and the other areas are black, and the appearance data image of different models is converted into a binary image, and the pixels with grayscale values ​​higher than a certain threshold in the appearance data image of different models are set to white, and the pixels with grayscale values ​​lower than the threshold are set to black;

[0020] Calculate the damaged area. In a binary image, white pixels correspond to the damaged area and black pixels correspond to other areas. Therefore, the number of white pixels in the image can be simply calculated to obtain the area of ​​the damaged area.

[0021] When converting to physical size, the area of ​​the damaged area is converted to physical size, and the actual area of ​​the damaged area is obtained by proportional conversion through camera calibration or the size of a known object;

[0022] Python

[0023] import cv2

[0024] #Read the image

[0025] image = cv2.imread('damaged_image.png',0)#Read as grayscale image

[0026] # Binarization

[0027] _,binary_image=cv2.threshold(image,127,255,cv2.THRESH_BINARY)

[0028] # Calculate the number of white pixels

[0029] damaged_area=cv2.countNonZero(binary_image)

[0030] print("Damaged area:", damaged_area)

[0031] #Optional: Convert area to physical size #pixel_to_mm = 0.1 #Assume each pixel represents 0.1 square millimeter #damaged_area_mm2 = damaged_area*pixel_to_mm*pixel_to_mm #print("Damage area:", damaged_area_mm2)

[0032] This allows the size or extent of the injury to be assessed.

[0033] As an optional solution to the technical solution of the present application document, the appearance model training of different vehicle models is established, and a convolutional neural network is selected for construction to extract features from the appearance images of different vehicle models and perform classification or detection. The selected appearance models of different vehicle models are trained using the labeled and enhanced appearance image data sets of different vehicle models, so that the appearance models of different vehicle models can accurately identify damaged or paint-chipped areas on the exterior of the car.

[0034] As an optional solution to the technical solution of the present application document, the appearance models of different vehicle models are optimized and tuned, and the model architecture is adjusted according to the analysis results. The adjustments include increasing / decreasing the number of layers, adjusting the number of nodes in each layer, using different activation functions, adding regularization layers, and using data enhancement technology in the training process to increase the diversity of training data and reduce the risk of overfitting. At the same time, regularization technology is used to reduce model complexity and improve generalization ability.

[0035] As an optional solution of the technical solution of this application document, the mobile detection device includes

[0036] An appearance inspection platform, a vehicle is arranged outside the appearance inspection platform;

[0037] A covering block, wherein the covering block is slidably disposed on the outside of the appearance inspection table through a moving assembly, and the vehicle is located on the inside of the covering block;

[0038] There are three detection heads, and two of them are slidably arranged on both sides of the cover block by extension drive to detect the bottom of the vehicle, and the other detection head is arranged on the inner side of the cover block by circumferential drive to detect the upper side of the vehicle.

[0039] 3. Beneficial effects

[0040] One or more technical solutions provided in the technical solution of this application have at least the following technical effects or advantages:

[0041] 1. This application can detect the damage of the vehicle and estimate the area of ​​the damage by setting up appearance models of different vehicle models, thereby improving the meticulousness and accuracy of the detection.

[0042] 2. Through the setting of the mobile components, the present application allows the covering block to be mobile and cover all around the outer side of the vehicle, thereby improving the coverage of the detection.

[0043] 3. Through the setting of the extension drive, the present application can drive the detection head on the extension block to carry out coverage detection on the bottom surface of the vehicle, thereby avoiding bumps on the bottom of the vehicle and certain safety hazards, and increasing the coverage of the detection range.

[0044] 4. This application uses a surround drive setting to allow the detection head on the cover block to move, so as to detect different positions and angles and improve the detection effect.

[0045] 5. This application uses the third motor to allow the detection head on the extension block to rotate in the reporting direction, so that the angle can be adjusted according to different situations to avoid blind spots or increase the range. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a schematic diagram of the structure of a vehicle appearance inspection system disclosed in a preferred embodiment of the present application;

[0047] Figure 2 This is a schematic diagram of the overall structure of a vehicle appearance inspection system disclosed in a preferred embodiment of the present application;

[0048] Figure 3 This is a structural disassembly diagram of a mobile component of a vehicle appearance inspection system disclosed in a preferred embodiment of the present application;

[0049] Figure 4 This is a split diagram of the extended drive structure of the automobile appearance inspection system disclosed in a preferred embodiment of the present application;

[0050] Figure 5 A cross-sectional view of the internal structure of an extension block of a vehicle appearance inspection system disclosed in a preferred embodiment of the present application;

[0051] Figure 6 An exploded diagram of a surround drive structure of a vehicle appearance inspection system disclosed in a preferred embodiment of the present application;

[0052] Explanation of the numbers in the figure: 1. Appearance inspection table; 2. Vehicle body; 3. Cover block; 31. Guide block; 32. Guide groove; 33. Screw rod; 34. Chassis A; 35. First motor; 4. Inspection head; 41. Limit cylinder; 42. Extension block; 43. Transmission rack; 44. Transmission gear; 45. Mechanical box A; 46. Chassis B; 47. Second motor; 401. Machine cavity; 402. Third motor; 4001. Surrounding block; 4002. Slide groove; 4003. Moving cavity; 4004. Driving gear; 4005. Incomplete internal gear; 4006. Mechanical box B; 4007. Fourth motor. DETAILED DESCRIPTION

[0053] The present application is further described in detail below in conjunction with the accompanying drawings.

[0054] Reference Figure 1 The embodiment of the present application provides a vehicle appearance detection system, comprising

[0055] Collect appearance data of different vehicle models, and collect data of different vehicle models and detection locations;

[0056] Preprocessing of appearance data of different models, such as denoising, for the convenience of subsequent model processing;

[0057] Establish appearance model training for different models, and use appearance data of different models to establish appearance models of different models and train them through appearance data of different models;

[0058] Optimize and tune the appearance models of different car models. Use the validation sets of appearance data of different car models to evaluate the performance of the trained appearance models of different car models on new data. Based on the evaluation results, adjust the hyperparameters to optimize the performance and generalization ability of the model.

[0059] Deployment of appearance models of different vehicle models: Deploy the trained appearance models of different vehicle models to actual application scenarios, acquire images through mobile cameras, and input the images into the appearance models of different vehicle models to detect damaged or peeling areas;

[0060] Mobile detection equipment will provide reminders of abnormal locations based on prediction results, and change locations for comprehensive testing.

[0061] By setting the appearance models of different vehicle models, the damage of the vehicle can be detected and the damaged area can be estimated, thereby improving the meticulousness and accuracy of the detection.

[0062] Reference Figure 1 and Figure 2 , the embodiment of the present application provides a vehicle appearance detection system, which collects appearance data of different vehicle models, and uses a detection head 4 to collect data;

[0063] The appearance image data of different car models are labeled, including intactness, degree of damage and model. At the same time, the damaged or paint-chipped areas in the image are marked.

[0064] Preprocessing of appearance data of different models. When the appearance data images of different models are not binary images, that is, the damaged areas are marked as white, while other areas are black, the appearance data images of different models are converted into binary images, and the pixels with grayscale values ​​higher than a certain threshold in the appearance data images of different models are set to white 255, and the pixels with grayscale values ​​lower than the threshold are set to black 0;

[0065] Calculate the damaged area. In a binary image, white pixels correspond to the damaged area and black pixels correspond to other areas. Therefore, the number of white pixels in the image can be simply calculated to obtain the area of ​​the damaged area.

[0066] When converting to physical size, the area of ​​the damaged area is converted to physical size, and the actual area of ​​the damaged area is obtained by proportional conversion through camera calibration or the size of a known object;

[0067] Python

[0068] import cv2

[0069] #Read the image

[0070] image = cv2.imread('damaged_image.png',0)#Read as grayscale image

[0071] # Binarization

[0072] _,binary_image=cv2.threshold(image,127,255,cv2.THRESH_BINARY)

[0073] # Calculate the number of white pixels and the area of ​​the damaged area

[0074] damaged_area=cv2.countNonZero(binary_image)

[0075] print("Number of pixels in the damaged area:", damaged_area)

[0076] #Optional: Convert area to physical size #pixel_to_mm = 0.1 #Assume each pixel represents 0.1 square millimeter #damaged_area_mm2 = damaged_area*pixel_to_mm*pixel_to_mm #print("Damage area square millimeter:", damaged_area_mm2)

[0077] This allows the size or extent of the injury to be assessed.

[0078] Establish training for appearance models of different car models, select convolutional neural networks for architecture, which are used to extract features from appearance images of different car models and perform classification or detection. Use labeled and enhanced appearance image datasets of different car models to train the selected appearance models of different car models, so that the appearance models of different car models can accurately identify damaged or paint-chipped areas on the car's appearance.

[0079] Optimize and tune the appearance models of different models, adjust the model architecture according to the analysis results, including increasing / decreasing the number of layers, adjusting the number of nodes in each layer, using different activation functions, adding regularization layers such as Dropout and L2 regularization, and using data enhancement technology during training to increase the diversity of training data and reduce the risk of overfitting. At the same time, use regularization techniques such as Dropout and L2 / L1 regularization to reduce model complexity and improve generalization ability.

[0080] Mobile testing equipment, including

[0081] An appearance inspection platform 1, a vehicle body 2 is arranged outside the appearance inspection platform 1;

[0082] A covering block 3, wherein the covering block 3 is slidably disposed on the outside of the appearance inspection platform 1 through a moving component, and the vehicle body 2 is located on the inside of the covering block 3;

[0083] There are three detection heads 4, and two of them are driven to slide on both sides of the cover block 3 by extension to detect the bottom of the vehicle body 2, and the other detection head 4 is driven to surround and set on the inner side of the cover block 3 to detect the upper side of the vehicle body 2.

[0084] Reference Figure 2 and Figure 3 The embodiment of the present application provides a vehicle appearance detection system, comprising

[0085] An appearance inspection platform 1, a vehicle body 2 is arranged outside the appearance inspection platform 1;

[0086] A covering block 3, wherein the covering block 3 is slidably disposed on the outside of the appearance inspection platform 1 through a moving component, and the vehicle body 2 is located on the inside of the covering block 3;

[0087] There are three detection heads 4, and two of them are driven to slide on both sides of the cover block 3 by extension to detect the bottom of the vehicle body 2, and the other detection head 4 is driven to surround and set on the inner side of the cover block 3 to detect the upper side of the vehicle body 2.

[0088] Reference Figure 3 , the embodiment of the present application provides a vehicle appearance inspection system, the cover block 3 is arranged in a ring structure;

[0089] The moving assembly includes a guide block 31 symmetrically fixedly arranged on the outer wall of the cover block 3, and the guide block 31 is slidably arranged inside a guide groove 32 opened on the outer wall of the appearance inspection platform 1;

[0090] A screw rod 33 is rotatably disposed inside one of the guide grooves 32 , and the screw rod 33 passes through the guide block 31 and is threadedly connected thereto;

[0091] A case A34 is embedded in one side of the guide groove 32 , and a first motor 35 is arranged inside the case A34 . The output shaft of the first motor 35 is coaxially fixedly connected to the lead screw 33 .

[0092] The rotation of the output shaft of the first motor 35 drives the screw rod 33 to rotate, and the screw rod 33 is threadedly connected to one of the guide blocks 31, so that the cover block 3 can perform mobile position detection. The screw rod 33 can improve the stability of movement and improve the movement accuracy. Through the setting of the moving component, the cover block 3 can be moved and covered all around the outer side of the vehicle, thereby improving the coverage of the detection.

[0093] Reference Figure 4, the embodiment of the present application provides a vehicle appearance inspection system, the extension drive includes two limit cylinders 41 symmetrically fixedly arranged at the outer end of the cover block 3, an extension block 42 is slidably arranged inside the limit cylinder 41, and the inner end of the extension block 42 is connected and fixed to the inspection head 4;

[0094] A transmission rack 43 is fixedly arranged on the outer wall of the extension block 42, and a transmission gear 44 is meshedly connected to the outer wall of the transmission rack 43. The transmission gear 44 is arranged inside a mechanical box A45 fixedly arranged on the outer wall of the limiting cylinder 41;

[0095] A machine box B46 is fixedly arranged on the outer wall of the machine box A45 , and a second motor 47 is arranged inside the machine box B46 . The output shaft of the second motor 47 is coaxially fixedly connected with the transmission gear 44 .

[0096] The output shaft of the second motor 47 rotates to drive the transmission gear 44, which is meshed with the transmission rack 43. At this time, the extension block 42 is driven to extend and retract, thereby reaching the bottom of the vehicle or moving away. Through the setting of the extension drive, the detection head 4 on the extension block 42 can be driven to perform coverage detection on the bottom of the vehicle, avoiding bumps on the bottom of the vehicle and certain safety hazards, thereby increasing the coverage of the detection range.

[0097] Reference Figure 5 The embodiment of the present application provides an automobile appearance inspection system, wherein an engine cavity 401 is provided inside the extension block 42 , a third motor 402 is arranged inside the engine cavity 401 , and an output shaft of the third motor 402 is connected and fixed to the outer wall of the inspection head 4 .

[0098] The detection head 4 is driven to rotate by the output shaft of the third motor 402 to perform detection at other angles. By setting the third motor 402, the detection head 4 on the extension block 42 can be rotated in a certain direction, so that the angle can be adjusted according to different situations to avoid blind spots or increase the range.

[0099] Reference Figure 6 , the embodiment of the present application provides a vehicle appearance inspection system, the surround drive includes a surround block 4001, and the surround block 4001 is connected and fixed to the outer wall of another inspection head 4;

[0100] The surrounding block 4001 is slidably disposed inside the slide groove 4002 provided on the inner wall of the covering block 3;

[0101] A moving cavity 4003 is provided inside the surrounding block 4001, and a driving gear 4004 is provided inside the moving cavity 4003. The driving gear 4004 is meshed with an incomplete internal gear 4005 fixedly provided on the inner wall of the slide groove 4002. A mechanical box B4006 is provided on the outer wall of the surrounding block 4001, and a fourth motor 4007 is provided inside the mechanical box B4006. The output shaft of the fourth motor 4007 is coaxially fixedly connected with the driving gear 4004.

[0102] The output shaft of the fourth motor 4007 rotates to drive the driving gear 4004, and the driving gear 4004 engages with the incomplete internal gear 4005. At this time, the surround block 4001 slides in the slide groove 4002, so that the detection head 4 on the covering block 3 moves along the contour of the covering block 3. Through the setting of the surround drive, the detection head 4 on the covering block 3 can be moved, so as to detect different positions and angles, thereby improving the detection effect.

[0103] When the automobile appearance inspection system is needed, first, data of different vehicle models and inspection positions are collected; the appearance image data of different vehicle models are annotated, including intact, damaged degree and model, and at the same time, the damaged or paint-chipped area in the image is marked. At this time, the appearance data of different vehicle models are pre-processed such as denoising to facilitate subsequent model processing; when the appearance data image of different vehicle models is not a binary image, that is, the damaged area has been marked as white, and other areas are black, the appearance data image of different vehicle models is converted into a binary image, and the pixels with grayscale values ​​higher than a certain threshold in the appearance data image of different vehicle models are set to white 255, and the pixels with grayscale values ​​lower than the threshold are set to black 0; the damaged area is calculated. In the binary image, white pixels correspond to the damaged area, and black pixels correspond to other areas. Therefore, the number of white pixels in the image can be simply calculated, that is, the area of ​​the damaged area. When converted to physical size, the area of ​​the damaged area is converted into physical size, and the proportion is converted through camera calibration or the size of the known object to obtain the actual area of ​​the damaged area;

[0104] Python

[0105] import cv2

[0106] #Read the image

[0107] image = cv2.imread('damaged_image.png',0)#Read as grayscale image

[0108] # Binarization

[0109] _,binary_image=cv2.threshold(image,127,255,cv2.THRESH_BINARY)

[0110] # Calculate the number of white pixels and the area of ​​the damaged area

[0111] damaged_area=cv2.countNonZero(binary_image)

[0112] print("Number of pixels in the damaged area:", damaged_area)

[0113] #Optional: Convert area to physical size #pixel_to_mm = 0.1 #Assume each pixel represents 0.1 square millimeter #damaged_area_mm2 = damaged_area*pixel_to_mm*pixel_to_mm #print("Damage area square millimeter:", damaged_area_mm2)

[0114] In order to assess the size or degree of damage, at this time, the appearance data of different models are used to establish appearance models of different models and train them through the appearance data of different models; a convolutional neural network is selected for construction to extract features from the appearance images of different models and classify or detect them, and the labeled and enhanced appearance image data sets of different models are used to train the selected appearance models of different models, so that the appearance models of different models can accurately identify the damage or paint peeling areas in the appearance of the car. At this time, the validation set of the appearance data of different models is used to evaluate the performance of the trained appearance models of different models on new data. According to the evaluation results, the hyperparameters are adjusted to optimize the performance and generalization ability of the model; the model architecture is adjusted according to the analysis results, including increasing / decreasing the number of layers, adjusting the number of nodes in each layer, using different activation functions, adding regularization layers such as Dropout and L2 regularization, using data enhancement technology in the training process to increase the diversity of training data and reduce the risk of overfitting. At the same time, regularization technologies such as Dropout and L2 / L1 regularization are used to reduce the complexity of the model and improve the generalization ability. At this time, the trained appearance models of different models are deployed to actual application scenarios, and images are acquired through mobile cameras. The image is taken and input into the appearance models of different vehicle models to detect the damaged or paint-shedding areas. At this time, the abnormal position is reminded according to the prediction results, and the position is changed for comprehensive detection, so that the first motor 35 output shaft rotates to drive the screw rod 33 to rotate, and the screw rod 33 is threadedly connected to one of the guide blocks 31, so that the cover block 3 can be moved. The screw rod 33 can improve the stability of movement and improve the movement accuracy. At the same time, the second motor 47 output shaft rotates to drive the transmission gear 44 to rotate, and the transmission gear 44 is meshed with the transmission rack 43, which will drive the extension block 42 to extend and retract. , so as to reach the bottom of the vehicle, or move it away, and cooperate with the rotation of the output shaft of the third motor 402 to drive the detection head 4 to rotate, and detect other angles. At the same time, the output shaft of the fourth motor 4007 drives the driving gear 4004 to rotate, and the driving gear 4004 is engaged with the incomplete internal gear 4005. At this time, the surrounding block 4001 is allowed to slide in the slide groove 4002, so that the detection head 4 on the covering block 3 moves along the contour of the covering block 3. Through the setting of the surrounding drive, the detection head 4 on the covering block 3 can be moved, so as to detect different positions and angles, thereby improving the detection effect.

Claims

1. A vehicle appearance inspection system, characterized in that: Include: Collect appearance data of different vehicle models, and collect data of different vehicle models and detection locations; Preprocessing of appearance data of different models, such as denoising, for the convenience of subsequent model processing; Establish appearance model training for different models, and use appearance data of different models to establish appearance models of different models and train them through appearance data of different models; Optimize and tune the appearance models of different car models. Use the validation sets of appearance data of different car models to evaluate the performance of the trained appearance models of different car models on new data. Based on the evaluation results, adjust the hyperparameters to optimize the performance and generalization ability of the model. Deployment of appearance models of different vehicle models: Deploy the trained appearance models of different vehicle models to actual application scenarios, acquire images through mobile cameras, and input the images into the appearance models of different vehicle models to detect damaged or peeling areas; Mobile detection equipment will provide reminders of abnormal locations based on prediction results, and change locations for comprehensive testing.

2. The automobile appearance inspection system according to claim 1, characterized in that: The data on the appearance of different vehicle models is collected by using a detection head (4); The appearance image data of different car models are labeled, including intactness, degree of damage and model. At the same time, the damaged or paint-chipped areas in the image are marked.

3. The automobile appearance inspection system according to claim 1, characterized in that: The preprocessing of the appearance data of different models is as follows: when the appearance data image of different models is not a binary image, that is, the damaged area has been marked as white, and the other areas are black, the appearance data image of different models is converted into a binary image, and the pixels with grayscale values ​​higher than a certain threshold in the appearance data image of different models are set to white (255), and the pixels with grayscale values ​​lower than the threshold are set to black (0); Calculate the damaged area. In a binary image, white pixels correspond to the damaged area and black pixels correspond to other areas. Therefore, the number of white pixels in the image can be simply calculated to obtain the area of ​​the damaged area. When converting to physical size, the area of ​​the damaged area is converted to physical size, and the actual area of ​​the damaged area is obtained by proportional conversion through camera calibration or the size of a known object; Python import cv2 #Read the image image=cv2.imread('damaged_image.png',0)#Read as grayscale image#binarization _,binary_image=cv2.threshold(image,127,255,cv2.THRESH_BINARY) #Calculate the number of white pixels (damaged area) damaged_area=cv2.countNonZero(binary_image) print("Damaged area (number of pixels):", damaged_area) #Optional: Convert the area to physical size #pixel_to_mm = 0.1 #Assume that each pixel represents 0.1 square millimeter #damaged_area_mm2 = damaged_area*pixel_to_mm*pixel_to_mm #print("Damaged area area (square millimeters) :", damaged_area_mm2) to assess the size or extent of the damage.

4. The automobile appearance inspection system according to claim 1, characterized in that: The training of appearance models of different car models is established, and a convolutional neural network is selected for construction to extract features from appearance images of different car models and perform classification or detection. The selected appearance models of different car models are trained using labeled and enhanced appearance image data sets of different car models, so that the appearance models of different car models can accurately identify damaged or paint-chipped areas on the car appearance.

5. The automobile appearance inspection system according to claim 1, characterized in that: The appearance models of different vehicle models are optimized and tuned, and the model architecture is adjusted according to the analysis results. The adjustments include increasing / decreasing the number of layers, adjusting the number of nodes in each layer, using different activation functions, adding regularization layers (such as Dropout and L2 regularization), and using data enhancement technology in the training process to increase the diversity of training data and reduce the risk of overfitting. At the same time, regularization technology (such as Dropout and L2 / L1 regularization) is used to reduce model complexity and improve generalization ability.

6. The automobile appearance inspection system according to claim 1, characterized in that: The mobile detection device comprises An appearance inspection platform (1), wherein a vehicle (2) is arranged outside the appearance inspection platform (1); A covering block (3), wherein the covering block (3) is slidably arranged on the outside of the appearance inspection platform (1) through a moving component, and the vehicle (2) is located on the inside of the covering block (3); The detection heads (4) are provided with three detection heads (4), and two of the detection heads (4) are slidably arranged on both sides of the cover block (3) by extension, so as to detect the bottom of the vehicle (2), and the other detection head (4) is arranged on the inner side of the cover block (3) by circumferential driving, so as to detect the upper side of the vehicle (2).

Citation Information

Patent Citations

  • Automobile appearance detection system and use method thereof

    CN115684172A

Cited By

  • Device and method for detecting defects on appearance surface of automobile

    CN121656280A