Vehicle damage detection method, storage medium, program product, controller and vehicle

Abstract

The invention relates to a vehicle damage detection method, a storage medium, a program product, a controller and a vehicle, and the vehicle damage detection method comprises the steps: obtaining a vehicle appearance image based on a visual sensor of the vehicle, and analyzing the vehicle appearance image to determine a damage detection result of the vehicle. According to the technical scheme, the high-definition image is captured in real time through the multi-path camera of the vehicle, and the image is automatically acquired and analyzed to identify the damage type / position / area, so that subjective experience judgment of personnel is replaced, and the vehicle damage detection efficiency is improved. The whole loss assessment process does not need getting-off operation of personnel, the vehicle evacuates immediately after detection, secondary accidents are avoided, and efficient and accurate all-weather automatic vehicle loss detection is achieved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a vehicle damage detection method, storage medium, program product, controller, and vehicle. Background Technology

[0002] With the increase in car ownership, road conditions are becoming more complex, and traffic accidents are gradually increasing. At this time, the intervention of car insurance and other related insurance services effectively protects the vehicle property safety of vehicle owners.

[0003] However, car insurance claims require damage assessment of the vehicle. Currently, vehicle damage assessment mainly relies on professional damage assessors conducting on-site inspections and judgments at the accident scene, subjectively judging the damage based on extensive experience accumulated through past experience and training. Furthermore, factors such as lighting conditions, shooting angle, and device resolution can interfere with the accuracy of images captured by mobile devices, making it difficult to fully represent the damage, often resulting in the omission of minor and hidden damage.

[0004] Therefore, this offline, on-site vehicle damage identification method has shortcomings such as high labor costs, low efficiency, strong subjectivity, and inaccurate identification. At the same time, during the period when the accident vehicle remains on the road for a long time while waiting for the damage assessor to arrive, there is a risk of secondary traffic accidents. Summary of the Invention

[0005] This application provides a vehicle damage detection method, storage medium, program product, controller, and vehicle, which improves the timeliness of vehicle damage detection and identification, thereby at least partially solving the above-mentioned technical problems.

[0006] To achieve the above objectives, according to a first aspect of this application, a vehicle damage detection method is provided, comprising: acquiring a vehicle exterior image based on the vehicle's own visual sensor, and analyzing the vehicle exterior image to determine the vehicle damage detection result.

[0007] Optionally, the method further includes:

[0008] In the event of a collision, images of the vehicle's exterior are acquired using the vehicle's own visual sensors.

[0009] Optionally, the method further includes determining that a vehicle collision event has occurred upon detecting a collision signal.

[0010] Optionally, acquiring vehicle exterior images based on the vehicle's own visual sensors includes:

[0011] Determine the collision location based on the collision signal;

[0012] The vehicle exterior image is acquired by the vehicle's visual sensor based on the collision location.

[0013] Optionally, analyzing the vehicle exterior images to determine the vehicle damage detection results includes:

[0014] At least one neural network is selected to detect the vehicle appearance image to obtain at least one vehicle appearance feature;

[0015] Based on the at least one vehicle appearance feature, the damage detection result of the vehicle is determined.

[0016] Optionally, the neural network includes a vehicle damage detection network, and the damage detection result includes a single-frame damage detection result, which includes at least one of damage type, damage location, and damage area.

[0017] Optionally, the neural network includes a vehicle temporal change detection network, which determines the vehicle damage detection result based on the at least one vehicle appearance feature, including:

[0018] Based on the vehicle's appearance features, the changes in the vehicle's appearance are obtained, and the temporal damage detection results of the vehicle are determined based on the changes in the vehicle's appearance.

[0019] The temporal damage detection results include at least one of the following: change type, change area, change intensity, and associated damage type.

[0020] Optionally, the vehicle appearance features include a first vehicle appearance feature corresponding to a historical reference frame and a second vehicle appearance feature corresponding to the current test frame. The step of obtaining changes in the vehicle appearance based on the vehicle appearance features includes:

[0021] By comparing the appearance features of the first vehicle with those of the second vehicle, the changes in the vehicle's appearance can be obtained.

[0022] The historical reference frame is an image of the vehicle in an undamaged state, and the current test frame is an image of the current vehicle appearance.

[0023] Optionally, the step of selecting at least one neural network to detect the vehicle exterior image to obtain at least one vehicle exterior feature includes:

[0024] Based on the image resource parameters corresponding to the vehicle appearance image, a target neural network is selected from at least one neural network;

[0025] The vehicle exterior image is detected using the target neural network to obtain vehicle exterior features;

[0026] The image resource parameters include at least one of image size, image duration, and the location of the visual sensor that acquires the vehicle exterior image.

[0027] Optionally, if the image size exceeds a preset resolution threshold, the image duration is less than a preset duration threshold, and / or the visual sensor is located at a position covering the side field of view of the vehicle body, the target neural network model is a vehicle damage detection network.

[0028] When the image size exceeds a preset resolution threshold, the image duration is not less than a preset duration threshold, and the visual sensor is located at a position covering the vehicle's forward field of view and / or covering the vehicle's rearward field of view, the target neural network model is a vehicle temporal change detection network.

[0029] Optionally, based on the at least one vehicle appearance feature, the damage detection result of the vehicle is determined, including:

[0030] Based on the appearance features of each vehicle, the first damage detection result corresponding to the appearance features of the vehicle is determined respectively;

[0031] And / or,

[0032] The damage detection result of the vehicle is determined by using a fusion model to perform feature fusion on the at least one first damage detection result;

[0033] The damage detection results of the vehicle include at least one of the following: comprehensive damage type, comprehensive damage location, comprehensive damage area, and damage confidence level.

[0034] Optionally, the first damage detection result includes single-frame damage detection results and temporal damage detection results. A fusion model is used to perform feature fusion on the at least one first damage detection result to determine the vehicle's damage detection result, including:

[0035] The single-frame damage detection results and temporal damage detection results are used as input features and input into the trained fusion model.

[0036] The fusion damage result is generated through the secondary decision model in the fusion model;

[0037] The secondary decision model is obtained by training and learning the fusion relationship between single-frame damage detection results and temporal damage detection results.

[0038] Optionally, after analyzing the vehicle exterior images to determine the damage detection results of the vehicle, the method further includes:

[0039] A vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results; wherein, the vehicle component detection results include at least one of component type, component location, and component area;

[0040] By combining the inspection results of the self-propelled vehicle components with the damage inspection results of the vehicle, a vehicle damage assessment result is generated;

[0041] The vehicle damage assessment results include at least one of the following: damage extent, damage type, damaged components, and repair estimate.

[0042] Optionally, the integration of the self-vehicle component inspection results with the vehicle damage inspection results includes:

[0043] The damaged location in the damage detection results is mapped to the corresponding component location area in the vehicle component detection results to identify the damaged component.

[0044] Optionally, vehicle damage assessment results are generated, including:

[0045] The percentage of the damaged area relative to the area of ​​the corresponding component in the damage detection results is calculated as the degree of damage.

[0046] Optionally, vehicle damage assessment results are generated, including:

[0047] By associating the damage type with the component type in the damage detection results, a damaged component identifier is generated.

[0048] Optionally, vehicle damage assessment results are generated, including:

[0049] The repair estimate is calculated based on the damaged component identification, the degree of damage, and the preset repair knowledge base.

[0050] Optionally, after generating the vehicle damage assessment result, the method further includes:

[0051] The vehicle damage assessment results are sent to the insurance claims and after-sales service server.

[0052] Optionally, in the event of a collision signal being detected, the method further includes:

[0053] A vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results; wherein, the vehicle component detection results include at least one of component type, component location, and component area;

[0054] The damage discrimination model is used to analyze the detection results of the vehicle parts in order to determine whether the corresponding vehicle parts are damaged.

[0055] Optionally, the detection results of the vehicle components are analyzed using a damage discrimination model, including:

[0056] Extract the image features of the vehicle component inspection results;

[0057] Damage presence is determined based on extracted image features, and the output is a probability value representing the likelihood of damage.

[0058] Optionally, determine whether the corresponding vehicle parts are damaged, including:

[0059] If the probability value is higher than a set threshold, the corresponding vehicle component is determined to be damaged.

[0060] If the probability value is not higher than the set threshold, the corresponding vehicle component is determined to be undamaged.

[0061] Optionally, if it is determined that the corresponding vehicle component is damaged, the method further includes:

[0062] The appearance image of damaged vehicle parts is detected and segmented using a vehicle damage segmentation network to obtain a second detection result; wherein, the second detection result includes at least one of the damaged part type, the damaged part location, and the damaged part area;

[0063] By combining the second detection result with the damage detection result of the vehicle, a vehicle damage assessment result is generated;

[0064] The damage assessment results for the vehicle include at least one of the following: the extent of damage, the type of damage, the damaged components, and the estimated repair cost.

[0065] Optionally, analyzing the vehicle exterior images to determine the vehicle damage detection results includes:

[0066] A vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results; wherein, the vehicle component detection results include at least one of component type, component location, and component area;

[0067] The detection network is used to detect and identify the detection results of the vehicle parts to determine the damage detection results of the vehicle.

[0068] The damage detection results of the vehicle include at least one of the damage type and the damage area.

[0069] Optionally, the vision sensor covers at least one of the vehicle's forward field of view, rearward field of view, and lateral field of view.

[0070] According to a second aspect of this application, a computer-readable storage medium is provided that stores a computer program or instructions thereon, which, when executed by a processor, implement the steps of any of the methods provided in the embodiments of this application.

[0071] According to a third aspect of this application, a computer program product is also provided, characterized in that it includes a computer program or instructions, which, when executed by a processor, implement the steps of any of the methods provided in the embodiments of this application.

[0072] According to a fourth aspect of this application, a controller is also provided, on which a computer program or instructions are stored, characterized in that, when the computer program or instructions are executed by a processor, they implement the steps of any of the methods provided in the embodiments of this application.

[0073] According to a fifth aspect of this application, a vehicle is provided, including the controller described above, or performing the steps of any of the methods provided in the embodiments of this application.

[0074] In summary, the embodiments of this application, through the above technical solution, utilize multi-channel cameras on the vehicle to capture high-definition images in real time, automatically acquire and analyze the images to identify the type, location, and area of ​​damage, thereby replacing subjective judgment by personnel and improving the efficiency of vehicle damage detection. The entire damage assessment process requires no personnel to leave the vehicle, and the vehicle is immediately removed after the inspection, avoiding secondary accidents and achieving efficient, accurate, and automated all-weather vehicle damage detection.

[0075] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0076] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0077] To gain a more complete understanding of this application and its beneficial effects, the following description will be provided in conjunction with the accompanying drawings, wherein the same reference numerals in the following description denote the same parts.

[0078] Figure 1 This is a flowchart illustrating the steps of a vehicle damage detection method provided in an exemplary embodiment of this application;

[0079] Figure 2 This is a schematic diagram of the vehicle body camera arrangement provided in an exemplary embodiment of this application;

[0080] Figure 3 This is a video image diagram illustrating the vehicle exterior image obtained in the vehicle damage detection method provided in the exemplary embodiments of this application;

[0081] Figure 4This is a logical schematic diagram of the vehicle damage detection method based on a self-vehicle damage detection network provided in the exemplary embodiments of this application;

[0082] Figure 5 This is a logical schematic diagram of the vehicle damage detection method based on a vehicle time-series change detection network provided in an exemplary embodiment of this application for vehicle damage detection;

[0083] Figure 6 This is a logical schematic diagram of the vehicle damage detection method based on a fusion model provided in the exemplary embodiments of this application;

[0084] Figure 7 This is a logical schematic diagram of the vehicle damage detection method based on a vehicle component detection network provided in an exemplary embodiment of this application.

[0085] Figure 8 This is a logical illustration of the vehicle damage detection method provided in the exemplary embodiments of this application. Figure 1 ;

[0086] Figure 9 This is a schematic diagram of a vehicle collision provided in an exemplary embodiment of this application;

[0087] Figure 10 This is a schematic diagram of vehicle image preprocessing in the vehicle damage detection method provided in the exemplary embodiment of this application;

[0088] Figure 11 This is a schematic diagram of the damage discrimination model executing the vehicle damage discrimination strategy in the vehicle damage detection method provided in the exemplary embodiment of this application;

[0089] Figure 12 This is a logical illustration of the vehicle damage detection method provided in the exemplary embodiments of this application. Figure 2 ;

[0090] Figure 13 This is a schematic diagram of the vehicle architecture provided in an exemplary embodiment of this application. Detailed Implementation

[0091] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0092] Based on the problems mentioned in the background technology, traditional vehicle damage assessment methods rely on professional damage assessors conducting on-site inspections and judgments. This assessment method often depends on the experience of the damage assessors to make subjective judgments about the damage, resulting in high labor costs, low timeliness, and strong subjectivity. Furthermore, while waiting for the damage assessors, the accident vehicle remains at the accident scene for an extended period, posing a risk of secondary traffic accidents. With the development of technology and the widespread application of smart mobile terminals, a smart damage assessment model using mobile terminals has emerged. In this model, the car owner uses a handheld mobile terminal (such as a mobile phone) to take several images of the vehicle's exterior. A preset recognition model then identifies components and damage in these images, and the damage assessment results determine whether the vehicle has external damage. However, this vehicle damage recognition model typically relies on the car owner holding their own mobile terminal (phone) to take images of the vehicle damage. The selection of the shooting location often depends on the car owner's subjective judgment. For some minor accidents that are not easily noticed or subtle damage that is not easily visible to the naked eye, the car owner may find it difficult to identify and assess the damage immediately. Meanwhile, this shooting operation requires personnel to walk around the vehicle to take images. On the one hand, accident scenes often involve heavy traffic or poor road conditions, which can easily cause secondary injuries to people. On the other hand, the images of the vehicle's exterior taken are easily affected by factors such as light intensity, shooting angle, and device resolution. The image quality of the images collected by the mobile terminal is difficult to guarantee, and it is difficult to fully present the damage. Minor and hidden damages are often missed, which will affect the accuracy of the recognition model in identifying vehicle damage and lead to deviations in the final vehicle damage detection results.

[0093] Therefore, to address the aforementioned problems, this application proposes a vehicle damage detection method, storage medium, program product, controller, and vehicle. The method utilizes the collaboration of an onboard vision sensor and a deep learning model to capture high-definition images in real time via multiple cameras on the vehicle. These images are then automatically acquired and analyzed by a dedicated neural network to automatically identify the type, location, and area of ​​damage, thereby replacing subjective judgment by personnel and improving the efficiency of vehicle damage detection. The entire damage assessment process requires no personnel to leave the vehicle, and the vehicle can be immediately removed after the inspection, preventing secondary accidents. By fusing component segmentation and temporal change detection, interference from the external environment on image capture and recognition can be avoided, improving the accuracy of vehicle damage detection.

[0094] This application provides a vehicle damage detection method. Please refer to [link / reference]. Figure 1 The vehicle damage detection method provided in this application includes step 100, which will be described in detail below.

[0095] Step 100: Acquire vehicle exterior images based on the vehicle's own vision sensors, and analyze the vehicle exterior images to determine the damage detection results of the vehicle.

[0096] In this embodiment, high-definition images are captured in real time by multiple cameras on the vehicle. The images are automatically acquired and analyzed to identify the type, location, and area of ​​damage, thereby replacing subjective judgment by personnel and improving the efficiency of vehicle damage detection. No personnel need to leave the vehicle during the entire damage assessment process, and the vehicle is immediately removed after the inspection, preventing secondary accidents.

[0097] In some embodiments, acquiring vehicle exterior images based on the vehicle's own vision sensors includes:

[0098] The vehicle's body video stream information is acquired in real time by its own multi-channel image sensors, and the video stream information is processed to obtain vehicle exterior images.

[0099] Optionally, the vehicle exterior image includes images of vehicle body components.

[0100] Please see Figure 2 , Figure 2 This is a schematic diagram of the vehicle body camera arrangement provided in an exemplary embodiment of this application.

[0101] The multiple image sensors are arranged to cover the vehicle's forward, rearward, and lateral fields of view.

[0102] In some embodiments, the camera covering the vehicle's forward field of view may optionally include at least one of a front-view main camera, a front-view narrow-angle camera, a front-view wide-angle camera, and a front-view fisheye camera, for acquiring high-quality images of the area directly in front of the vehicle and the front bumper, hood, front fender, etc.

[0103] In some embodiments, the camera covering the rearward view of the vehicle may optionally include at least one of a rearview camera and a rear surround fisheye camera, for acquiring images of the area directly behind the vehicle and the rear bumper, trunk lid, rear fender, etc.

[0104] In some embodiments, the camera covering the side view of the vehicle may optionally include at least one of a side surround view fisheye camera (typically located below the left and right rearview mirrors), a side front camera (typically located on both sides of the left and right front fenders or front bumper), and a side rear camera (typically located on both sides of the left and right rear fenders or rear bumper), for acquiring images of the left and right sides of the vehicle body (including doors, side skirts, wheel arches, and the sides of the front and rear fenders).

[0105] Optionally, the vehicle's own multimedia cameras also include all multimedia cameras visible from the vehicle body.

[0106] It should be noted that the core criterion for selecting a multimedia camera is whether its FOV (Field of View) can effectively cover the corresponding area of ​​the vehicle body. The vehicle damage detection method provided in this application, which determines the information input source for vehicle damage detection results, is limited to multimedia cameras that can actually capture images of the vehicle body surface. The specific cameras selected depend on the actual multimedia camera hardware configuration of different vehicle models and their respective FOV coverage capabilities.

[0107] In some embodiments, the vehicle body video stream information is acquired in real time by the vehicle's own multi-channel image sensors, the video stream information is preprocessed, and the vehicle body image is extracted based on the preprocessed video stream information.

[0108] Specifically, the vehicle body video stream information is preprocessed, and based on the preprocessed video stream information, an image of the vehicle body is obtained through image extraction technology. The principle of the preprocessing operation can refer to conventional techniques, such as filtering, denoising, and enhancing the video stream information to reduce noise and improve image quality, which will not be elaborated in this application.

[0109] Among these methods, the vehicle body image is obtained through image extraction technology based on the preprocessed video stream information, optionally including:

[0110] A predefined fixed mask template is used to perform the cropping operation of the vehicle body ROI (Region of Interest), which is generated based on the vehicle CAD model;

[0111] Images of vehicle body parts are obtained by extracting vehicle body regions using a dynamic segmentation network, where the dynamic segmentation network is a semantic segmentation model.

[0112] Please see Figure 3 , Figure 3 This is a schematic diagram of a video frame used to acquire an image of a vehicle's exterior, provided in an exemplary embodiment of this application. Taking a side-view video frame as an example, the video frame image undergoes preprocessing to reduce noise and improve image quality. Subsequently, the preprocessed video frame image is processed by cropping the ROI region of the vehicle body area and applying a predefined fixed mask template (or vehicle damage segmentation network) to extract images of the vehicle body components.

[0113] In this embodiment, preprocessing of video stream information, cropping of ROI regions in the vehicle body area, using predefined fixed mask templates, or vehicle damage segmentation networks to extract images of vehicle body components can reduce the useless consumption of onboard computing power by non-target areas (environmental information such as road surface and driving that does not belong to the vehicle structure) and reduce the interference of complex environments on the identification of vehicle components and vehicle damage.

[0114] In some embodiments, analyzing the vehicle exterior images to determine the vehicle damage detection results includes:

[0115] At least one neural network is selected to detect the vehicle appearance image to obtain at least one vehicle appearance feature;

[0116] Based on the at least one vehicle appearance feature, the damage detection result of the vehicle is determined.

[0117] In some embodiments, at least one neural network is selected to detect the vehicle appearance image to obtain at least one vehicle appearance feature, including:

[0118] Based on the image resource parameters corresponding to the vehicle appearance image, a target neural network is selected from at least one neural network;

[0119] The vehicle exterior image is detected using the target neural network to obtain vehicle exterior features;

[0120] The image resource parameters include at least one of image size, image duration, and the location of the visual sensor that acquires the vehicle exterior image.

[0121] In some embodiments, when the image size exceeds a preset resolution threshold, the image duration is less than a preset duration threshold, and / or the visual sensor is located at a position covering the side view of the vehicle body, the target neural network model is a vehicle damage detection network.

[0122] Please see Figure 4 , Figure 4 This is a logical schematic diagram of the vehicle damage detection method based on a self-vehicle damage detection network provided in an exemplary embodiment of this application.

[0123] In some embodiments, the neural network includes a vehicle damage detection network, and the damage detection results include single-frame damage detection results.

[0124] Optionally, the single-frame damage detection result includes at least one of damage type, damage location, and damage area.

[0125] Damage type is an automatic classification based on the texture, shape, and depth features of the damaged area. For example, when the shape feature is a thin strip and the texture feature is a linear high-frequency edge, the corresponding damage type is scratch.

[0126] The location of the damage can be represented by a relative coordinate system or an absolute position mapping to indicate the specific location of the damage on the vehicle body. For example, a relative coordinate system is a two-dimensional grid coordinate system with the center point of the vehicle body as the origin. The specific location of the damage is represented by the coordinates of the damage point in this two-dimensional coordinate system.

[0127] The damaged area can be calculated by counting the number of effective pixels in the damaged mask.

[0128] In some embodiments, the vehicle damage detection network employs an encoder-decoder convolutional neural network.

[0129] Specifically, the encoder-decoder convolutional neural network consists of the following components:

[0130] Convolutional layers: extract local damage features (such as scratch textures and indentation gradients);

[0131] Pooling layer: Max pooling is used for dimensionality reduction to expand the receptive field;

[0132] Attention mechanism: Focusing on injury-sensitive areas.

[0133] Optionally, the structure of the vehicle damage detection network may include, but is not limited to, U-Net series, YOLO-SEG series, Mask R-CNN series, SegNet, PSPNet, SAM series, etc.

[0134] In some embodiments, taking a vehicle damage scenario as an example, the process of vehicle damage detection based on damage detection is explained. The vehicle damage detection network first scans the vehicle image through convolutional layers in the encoder stage to identify the basic features of the damage. For example, scratches appear as fine, elongated stripes, indentations form areas of localized light and dark variation, and cracks present as irregular linear structures. After each round of feature extraction, pooling layers compress the data size, expanding the field of view to capture a wider range of damage-related features. To address the metallic reflective interference unique to the vehicle environment, the network integrates a channel attention mechanism, automatically reducing the weight of bright areas while enhancing the response intensity of damage features in dark areas. Low-resolution feature maps are gradually restored to their original size through upsampling operations. During this process, the vehicle damage detection network achieves simultaneous multi-scale damage detection by fusing high-resolution details (such as micro-scratches) from the encoder stage and high-level semantic features (such as large-area indented areas) from the decoder. To accurately locate complex damage at sheet metal seams (such as dents on the edge of a car door), the network introduces a boundary refinement module. By comparing the brightness and gradient relationship of adjacent pixels, it eliminates jagged artifacts and outputs a smooth and accurate damage contour.

[0135] The images captured by multiple cameras on a vehicle (such as a front-view 4K main camera and a side-view fisheye camera) vary significantly in size. High-resolution images (e.g., 4096×2160 pixels) contain several times the pixel data of low-resolution images (e.g., 640×480 pixels). Furthermore, image duration is directly related to computational power consumption. When the image duration exceeds a preset threshold, or when the detected image size exceeds a preset threshold (e.g., resolution > 1920×1080), a vehicle damage detection network is used to detect the vehicle's exterior image. Only a single frame is detected and identified, avoiding computational waste and slow response times.

[0136] In some embodiments, when the image size exceeds a preset resolution threshold, the image duration is not less than a preset duration threshold, and the visual sensor is located at a position covering the forward field of view and / or the rear field of view of the vehicle, the target neural network model is a vehicle temporal change detection network.

[0137] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating the logic of a vehicle damage detection method based on a vehicle temporal change detection network provided in an exemplary embodiment of this application. In some embodiments, the neural network includes a vehicle temporal change detection network, which determines the vehicle damage detection result based on the at least one vehicle appearance feature, including:

[0138] Based on the vehicle's appearance features, the changes in the vehicle's appearance are obtained, and the temporal damage detection results of the vehicle are determined based on the changes in the vehicle's appearance.

[0139] The temporal damage detection results include at least one of the following: change type, change area, change intensity, and associated damage type.

[0140] The change type represents the dynamic evolution of classified damage based on the difference between historical reference frames and the current reference frame. For example, new damage (such as a dent that appears at the moment of collision) and damage expansion (such as the increase in the area of ​​the rusted region).

[0141] A change region is a pixel-level spatial representation of damage changes, such as heatmap coordinates (e.g., marker changes) or bounding boxes (tightly enclosing areas of the change region).

[0142] The intensity of change is a measure of the physical significance of damage changes, such as slight changes (paint oxidation) and drastic changes (sheet metal deformation).

[0143] In some embodiments, the vehicle appearance features include a first vehicle appearance feature corresponding to a historical reference frame and a second vehicle appearance feature corresponding to the current test frame. The step of obtaining changes in the vehicle appearance based on the vehicle appearance features includes:

[0144] By comparing the appearance features of the first vehicle with those of the second vehicle, the changes in the vehicle's appearance are obtained.

[0145] The historical reference frame is an image of the vehicle in an undamaged state, and the current test frame is an image of the current vehicle appearance.

[0146] After a vehicle leaves the factory or undergoes repair, an image of the vehicle's complete state is automatically captured and stored in the vehicle's infotainment system as a historical reference frame. When the vehicle's time-series change detection is activated, the current vehicle exterior image is used as the current test frame.

[0147] In some embodiments, different level features include:

[0148] Shallow features characterize the captured detailed textures (such as the linear edges of paint scratches);

[0149] Mid-level features characterize the identified structural contours (such as the curvature of a car door surface);

[0150] Deep features represent global features (such as the layout of vehicle body components).

[0151] Optionally, the backbone network includes at least one of ResNet, VGGNet, and Transformer.

[0152] Alternatively, the vehicle timing change detection network can be ChangeFormer, LightCDNet, ChangeNet, etc.

[0153] In this embodiment, the vehicle time-series change detection network can capture the complex patterns and dependencies of time-series data, and the vehicle body has better adaptability under different lighting and contrast environments.

[0154] Because the front of a vehicle has a high probability of damage in a frontal collision, and the collision may cause large-area sheet metal deformation or fragmentation, the embodiments of this application employ a temporal processing mode for the image sequence captured by the forward-facing camera. Specifically, multiple frames of images before and after the collision are continuously acquired (e.g., 5 frames / second), inter-frame motion vectors are extracted through optical flow estimation, and the deformation trend of the damaged area is detected using a deep learning model. Temporal features are utilized to enhance the ability to identify progressive damage such as the propagation of minor cracks and paint peeling. For reversing collisions or rear-end collisions, the rear-facing camera also employs a temporal processing mode, focusing on the damage evolution process of vulnerable parts such as the rear bumper and taillights. Since the probability of side collisions is relatively low and they are mostly instantaneous impacts, the embodiments of this application employ a single-frame processing mode for the images captured by the side-facing camera.

[0155] Please see Figure 6 , Figure 6This is a logical schematic diagram of a vehicle damage detection method based on a fusion model provided in an exemplary embodiment of this application. In some embodiments, determining the vehicle damage detection result based on the appearance features of the at least one vehicle includes:

[0156] Based on the appearance features of each vehicle, the first damage detection result corresponding to the appearance features of the vehicle is determined respectively;

[0157] And / or,

[0158] The damage detection result of the vehicle is determined by using a fusion model to perform feature fusion on the at least one first damage detection result.

[0159] The first damage detection result includes single-frame damage detection results and temporal damage detection results.

[0160] It is understandable that the vehicle appearance image is detected by the self-vehicle damage detection network to obtain the single-frame damage detection result of the vehicle, and the vehicle appearance image is detected by the self-vehicle temporal change detection network to obtain the changes in the vehicle appearance. The temporal damage detection result of the vehicle is determined based on the changes in the vehicle appearance. For specific implementation methods and more details, please refer to the previous embodiments, which will not be repeated here.

[0161] The vehicle damage detection results include at least one of the following: comprehensive damage type, comprehensive damage location, comprehensive damage area, and damage confidence level.

[0162] The comprehensive damage type is a damage classification result generated by fusing the damage type in the single-frame damage detection result and the associated damage type in the temporal damage detection result.

[0163] The comprehensive damage location is a spatial distribution result of damage generated by fusing the damage location in the single-frame damage detection result and the change region in the temporal damage detection result.

[0164] The overall damage area is a damage quantification result generated by fusing the damage area in the single-frame damage detection result and the change intensity in the temporal damage detection result.

[0165] Damage confidence is a reliability assessment value generated based on the degree of consistency between single-frame damage detection results and temporal damage detection results.

[0166] Optionally, the fusion model may include weighted methods, stacking strategies, etc.

[0167] In some embodiments, determining the fused damage result of the vehicle based on the single-frame damage detection result and the temporal damage detection result includes:

[0168] The single-frame damage detection results and temporal damage detection results are used as input features and input into the trained fusion model.

[0169] The fusion damage result is generated through the secondary decision model in the fusion model;

[0170] The secondary decision model is obtained by training and learning the fusion relationship between single-frame damage detection results and temporal damage detection results.

[0171] In this embodiment, vehicle damage detection is achieved through a fusion scheme based on a self-vehicle damage detection network and a temporal damage detection network, or either one. The self-vehicle damage detection network constructs features at the spatial level to detect vehicle damage, while the temporal damage detection network constructs features at the temporal level to detect changes in vehicle appearance. Combining the features obtained from the two detection networks can synergistically leverage the advantages of single-frame detail recognition and temporal anti-interference, accurately determining the vehicle damage detection results, and is not affected by the subjective factors of the damage assessor. The captured vehicle damage images have scale consistency and spatiotemporal consistency.

[0172] Please see Figure 7 , Figure 7 This is a logical schematic diagram of the vehicle damage detection method based on a vehicle component detection network provided in an exemplary embodiment of this application.

[0173] In some embodiments, when a collision signal is detected, the method further includes:

[0174] A deep learning-based vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results.

[0175] By spatially correlating and fusing the detection results of the self-vehicle components with the damage detection results of the vehicle, a vehicle damage assessment result is generated.

[0176] In some embodiments, the vehicle component detection network employs an encoder-decoder convolutional neural network.

[0177] Optionally, the inspection results of the vehicle parts may include at least one of the following: the type of the vehicle exterior parts, the location of the parts, and the area of ​​the parts.

[0178] In some embodiments, taking the lateral position of the vehicle as an example, the relative positions of the vehicle body surface and the on-board camera sensors in each direction are fixed. The working principle of the vehicle component detection network using deep learning to detect and segment the vehicle exterior image to obtain the vehicle component detection result is explained. When the vehicle is normally ignited and pneumatically operated, the multi-channel image sensors of the vehicle body capture and acquire video stream information of the lateral position of the vehicle body in real time, and input it into the vehicle component detection network. After the above-mentioned vehicle image preprocessing, the preprocessed lateral position exterior image video frame of the vehicle is obtained. When the preprocessed video frame is input into the convolutional neural network with an encoder-decoder structure, the encoder extracts the abstract features of the image layer by layer and focuses on the key component areas. The decoder restores the feature image to the original resolution, fuses the deep and shallow layer features, and outputs a smooth and accurate vehicle component detection result.

[0179] It is understandable that a deep learning-based vehicle component detection network, constructed using an encoder-decoder convolutional neural network, is used to detect and segment vehicle exterior images to obtain vehicle component detection results. For specific implementation methods and more details, please refer to the previous embodiments, which will not be repeated here.

[0180] Optionally, the damage assessment results for the vehicle include, but are not limited to, the extent of damage, the type of damage, the damaged components, and the estimated repair cost.

[0181] The degree of damage indicates the percentage of the area damaged relative to the area of ​​the associated component. For example, the damaged area of ​​the left front door is 1250 mm². 2 The total area of ​​the car doors is 25000 mm. 2 If so, the degree of damage is 5% (moderate damage).

[0182] Damage type is an automatic classification based on the texture, shape, and depth features of the damaged area. For example, when the shape feature is a thin strip and the texture feature is a linear high-frequency edge, the corresponding damage type is scratch.

[0183] Damaged parts are identified by analyzing vehicle exterior images to determine the overlap between the damage mask and the part's grinding process, thus identifying the part to which it belongs.

[0184] The repair estimate is the amount estimated by combining the damaged parts, the extent of the damage, and the costs of labor, materials, etc.

[0185] In some embodiments, analyzing the vehicle exterior images to determine the vehicle damage detection results includes:

[0186] A vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results.

[0187] The detection network is used to detect and identify the detection results of the vehicle parts to determine the damage detection results of the vehicle.

[0188] The vehicle component inspection results include at least one of component type, component location, and component area; the vehicle damage inspection results include at least one of damage type and damage area.

[0189] Specifically, the detection network can be a vehicle damage detection network, a vehicle time-series detection network, or a fusion model to detect and identify the detection results of the vehicle parts. For specific implementation methods and more details, please refer to the previous embodiments, which will not be repeated here.

[0190] Please see Figure 8 , Figure 8 This is a logical illustration of the vehicle damage detection method provided in the exemplary embodiments of this application. Figure 1 In some embodiments, the detection results of the vehicle components are spatially correlated with the damage detection results of the vehicle, including:

[0191] The damaged location in the damage detection results is mapped to the corresponding component location area in the vehicle component detection results to identify the damaged component.

[0192] In some embodiments, generating vehicle damage assessment results includes:

[0193] The percentage of the damaged area relative to the area of ​​the corresponding component in the damage detection results is calculated as the degree of damage.

[0194] In some embodiments, generating vehicle damage assessment results includes:

[0195] By associating the damage type with the component type in the damage detection results, a damaged component identifier is generated.

[0196] In some embodiments, generating vehicle damage assessment results includes:

[0197] The repair estimate is calculated based on the damaged component identification, the degree of damage, and the preset repair knowledge base.

[0198] It is understandable that by obtaining vehicle component inspection results through the vehicle component inspection network, including information such as component type, component location, and component area, and by obtaining single-frame damage detection results, temporal damage detection results, and / or fusing damage results through the vehicle damage detection network, including information such as damage type, damage area, and damage location, and by organically fusing the above information, including but not limited to the area ratio, overlap ratio, and positional relationship of the damaged parts on the exterior components, the vehicle damage assessment results, such as the degree of vehicle damage, damage type, and repair estimate, can be obtained.

[0199] In some embodiments, after using a deep learning-based vehicle component detection network to detect and segment vehicle exterior images to obtain vehicle component detection results, and combining the vehicle component detection results with the vehicle damage detection results to obtain the vehicle damage assessment result, the method further includes:

[0200] The damage assessment results for the vehicle will be sent to the insurance company's server.

[0201] The insurance company's server receives the vehicle damage assessment results and uses this information to recommend and allocate insurance claims and after-sales services.

[0202] In some embodiments, the method further includes acquiring an image of the vehicle's exterior based on the vehicle's own visual sensors in the event of a collision.

[0203] In some embodiments, the method further includes determining that a vehicle collision event has occurred upon detecting a collision signal.

[0204] In some embodiments, acquiring vehicle exterior images based on the vehicle's own visual sensors includes:

[0205] Determine the location of the collision event based on the collision signal; and

[0206] The vehicle's exterior image is obtained based on the vehicle's own visual sensors corresponding to the location where the collision event occurred.

[0207] Please see Figure 9 , Figure 9 This is a schematic diagram of a vehicle collision provided in an exemplary embodiment of this application. When an obstacle collides with the vehicle, the collision detection device installed on the vehicle generates a collision signal and determines the collision location based on the collision signal. It then collects video frames acquired by each visual sensor corresponding to the collision location as the basis for vehicle damage detection results.

[0208] Similarly, the vehicle's own visual sensors, which acquire the vehicle's video stream information corresponding to the location of the collision event, are preprocessed, and the vehicle's exterior image is obtained through image extraction technology based on the preprocessed video stream information.

[0209] Please see Figure 10 , Figure 10 This is a schematic diagram of vehicle image preprocessing provided in an exemplary embodiment of this application. Taking a side-view video frame of the vehicle as an example, the video frame image is preprocessed, ROI region cropped, and subjected to predefined fixed mask template operations to extract the target vehicle exterior image, including but not limited to images of various vehicle body parts and the category of the parts. Specific implementation methods and more details can be found in the preceding embodiments, and will not be repeated here.

[0210] In the embodiments of this application, when a vehicle collision occurs, the vehicle's exterior image is acquired by the vehicle's visual sensor corresponding to the collision location, and the vehicle's exterior image is preprocessed and the vehicle's components are segmented. This can reduce the useless consumption of onboard computing power in non-vehicle areas and reduce interference with vehicle damage identification in complex environments.

[0211] Please see Figure 11 , Figure 11 This is a schematic diagram of the damage model's judgment strategy for the detection results of vehicle parts in the vehicle damage detection method provided in the exemplary embodiment of this application.

[0212] The detection results of the vehicle parts are analyzed using a pre-trained damage discrimination model to determine whether the corresponding vehicle parts are damaged.

[0213] The analysis of the detection results of the vehicle components using a pre-trained damage discrimination model includes:

[0214] Extracting image features from the inspection results of vehicle parts;

[0215] Damage presence is determined based on extracted image features, and a probability value representing the likelihood of damage is output.

[0216] If the probability value is higher than a set threshold, the corresponding vehicle component is determined to be damaged.

[0217] If the probability value is not higher than the set threshold, it is determined that the corresponding vehicle component is not damaged.

[0218] In some embodiments, after analyzing the detection results of the vehicle parts using a pre-trained damage discrimination model to determine whether the corresponding vehicle parts are damaged, the method further includes:

[0219] Damage detection and analysis are selectively performed based on the judgment results.

[0220] Specifically, the analysis of vehicle exterior images is performed on the vehicle parts that are determined to be damaged to determine the damage detection results;

[0221] For vehicle parts that are determined to be undamaged, the analysis of the vehicle exterior images is not performed to determine the damage detection results.

[0222] Optionally, the damage discrimination model includes, but is not limited to, a binary classification network model, which analyzes the image features of the component and outputs a probability value (range 0-1) representing the likelihood of damage. When the probability value is higher than a set threshold (e.g., 0.5), it is determined to be damaged; otherwise, it is determined to be undamaged. In a preferred embodiment, this threshold can be dynamically adjusted according to safety requirements.

[0223] Through the pre-screening mechanism of the damage discrimination model, the system can reduce the detection calculations for obviously undamaged parts.

[0224] In some embodiments, the damage discrimination model is a deep convolutional neural network model;

[0225] The deep convolutional neural network model includes:

[0226] The input layer receives 3D structural images of vehicle body parts.

[0227] At least three convolutional layers, each containing multiple feature extraction units;

[0228] The output layer generates binary classification decision results pointing to the "normal" or "damaged" state;

[0229] The feature extraction unit abstracts visual features step by step through hierarchical connections, and the final decision result activates the "damaged" or "normal" output branch through threshold comparison.

[0230] The damage discrimination model adopts a deep convolutional network architecture, which learns the structural features of three-dimensional parts step by step through multiple feature extraction units, and finally outputs a deterministic classification result of "normal" or "damaged".

[0231] Please see Figure 12 , Figure 12 This is a schematic diagram of the damage discrimination model executing the vehicle damage discrimination strategy in the vehicle damage detection method provided in the exemplary embodiments of this application. Figure 2 As shown in the figure, (left) the input images of vehicle body parts are augmented to form a training sample set; (right) the damage discrimination model topology, whose training method specifically includes: collecting a large number of labeled images of vehicle body parts as training sample sets, with the labeling content including two real states: vehicle damage and no vehicle damage; performing data augmentation operations to improve the robustness of the model, using a deep convolutional neural network as the basic architecture, including but not limited to: residual learning modules of the ResNet series (such as ResNet-50 / 101); continuous convolutional layer design of the VGG series (such as VGG-16 / 19); multi-head self-attention mechanism of the Transformer architecture; learning discriminative features through end-to-end training, automatically extracting multi-level visual features (edge ​​→ texture → semantics) from the images; and establishing a mapping relationship between features and vehicle damage state.

[0232] In some embodiments, when it is determined that a corresponding vehicle component is damaged, the method further includes:

[0233] A deep learning-based vehicle damage segmentation network is used to detect and segment the appearance image of damaged vehicle parts to obtain a second detection result; wherein, the second detection result includes at least one of the damaged part type, the damaged part location, and the damaged part area;

[0234] By spatially correlating and fusing the second detection result with the vehicle damage detection result, a vehicle damage assessment result is generated;

[0235] The damage assessment results for the vehicle include at least one of the following: the extent of damage, the type of damage, the damaged components, and the estimated repair cost.

[0236] It should be noted that the above description of the process is for illustrative purposes only and does not limit the scope of this specification. Those skilled in the art can make various modifications and changes to the process under the guidance of this specification. However, these modifications and changes remain within the scope of this specification.

[0237] One or more embodiments of this application also provide a controller for executing the steps in the embodiments corresponding to the above-described vehicle damage detection method. For specific implementation methods and more details, please refer to the corresponding method section, which will not be repeated here.

[0238] For specific implementation examples of the above operations, please refer to the previous examples, which will not be repeated here.

[0239] In some embodiments, the controller may be a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor, or other electronic components.

[0240] This application also provides a computer-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the aforementioned vehicle damage detection method.

[0241] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0242] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0243] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0244] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0245] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0246] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0247] Computer-readable media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient media, such as modulated communication signals and carrier waves.

[0248] like Figure 13 The diagram shown is a schematic representation of a vehicle architecture provided in an embodiment of this application. In this embodiment, the vehicle includes a multimedia camera mounted on the vehicle body and a controller provided in any of the above embodiments. The controller is used to execute the vehicle damage detection method provided in any of the above embodiments. Alternatively, it executes the steps of the vehicle damage detection method provided in the embodiments of this application. In this embodiment, the vehicle can be a gasoline-powered vehicle, a plug-in hybrid electric vehicle, or a new energy vehicle, etc., and this disclosure does not specifically limit it.

[0249] In one embodiment, the vehicle can be configured for fully or partially autonomous driving. For example, the vehicle can control itself while in autonomous driving mode, and can determine the current state of the vehicle and its surrounding environment through human intervention, determine the possible behaviors of at least one other vehicle in the surrounding environment, and determine the confidence level corresponding to the probability of that other vehicle performing a possible behavior, and control the vehicle based on the determined information. When the vehicle is in autonomous driving mode, it can be configured to operate without human interaction.

[0250] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0251] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0252] The embodiments, implementation methods, and related technical features of this application can be combined and substituted for each other without conflict.

[0253] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application without departing from the scope of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for detecting vehicle damage, characterized in that, include: The vehicle's exterior images are acquired using the vehicle's own vision sensors, and the images are analyzed to determine the damage detection results.

2. The method according to claim 1, characterized in that, The method further includes: In the event of a collision, images of the vehicle's exterior are acquired using the vehicle's own visual sensors.

3. The method according to claim 2, characterized in that, The method also includes determining that a vehicle collision event has occurred upon detecting a collision signal.

4. The method according to claim 3, characterized in that, The acquisition of vehicle exterior images based on the vehicle's own visual sensors includes: Determine the collision location based on the collision signal; Based on the collision location, the vehicle's exterior image is acquired using the corresponding vehicle vision sensor.

5. The method according to claim 1, characterized in that, Analyzing the vehicle exterior images to determine the vehicle damage detection results includes: At least one neural network is selected to detect the vehicle appearance image to obtain at least one vehicle appearance feature; Based on the at least one vehicle appearance feature, the damage detection result of the vehicle is determined.

6. The method according to claim 5, characterized in that, The neural network includes a vehicle damage detection network, and the damage detection result includes a single-frame damage detection result, which includes at least one of damage type, damage location, and damage area.

7. The method according to claim 5, characterized in that, The neural network includes a vehicle temporal change detection network, which determines the vehicle damage detection result based on the at least one vehicle appearance feature, including: Based on the vehicle's appearance features, the changes in the vehicle's appearance are obtained, and the temporal damage detection results of the vehicle are determined based on the changes in the vehicle's appearance. The temporal damage detection results include at least one of the following: change type, change area, change intensity, and associated damage type.

8. The method according to claim 7, characterized in that, The vehicle appearance features include a first vehicle appearance feature corresponding to a historical reference frame and a second vehicle appearance feature corresponding to the current test frame. The step of obtaining changes in the vehicle appearance based on these features includes: By comparing the appearance features of the first vehicle with those of the second vehicle, the changes in the vehicle's appearance can be obtained. The historical reference frame is an image of the vehicle in an undamaged state, and the current test frame is an image of the current vehicle appearance.

9. The method according to claim 5, characterized in that, The step of selecting at least one neural network to detect the vehicle exterior image and obtaining at least one vehicle exterior feature includes: Based on the image resource parameters corresponding to the vehicle appearance image, a target neural network is selected from at least one neural network; The vehicle exterior image is detected using the target neural network to obtain vehicle exterior features; The image resource parameters include at least one of image size, image duration, and the location of the visual sensor that acquires the vehicle exterior image.

10. The method according to claim 9, characterized in that, When the image size exceeds a preset resolution threshold, the image duration is less than a preset duration threshold, and / or the visual sensor is located at a position covering the side field of view of the vehicle body, the target neural network model is a vehicle damage detection network; When the image size exceeds a preset resolution threshold, the image duration is not less than a preset duration threshold, and the visual sensor is located at a position covering the vehicle's forward field of view and / or covering the vehicle's rearward field of view, the target neural network model is a vehicle temporal change detection network.

11. The method according to any one of claims 6 to 10, characterized in that, Based on the at least one vehicle appearance feature, the damage detection result of the vehicle is determined, including: Based on the appearance features of each vehicle, the first damage detection result corresponding to the appearance features of the vehicle is determined respectively; And / or, The damage detection result of the vehicle is determined by using a fusion model to perform feature fusion on the at least one first damage detection result; The damage detection results of the vehicle include at least one of the following: comprehensive damage type, comprehensive damage location, comprehensive damage area, and damage confidence level.

12. The method according to claim 11, characterized in that, The first damage detection result includes single-frame damage detection result and temporal damage detection result. A fusion model is used to perform feature fusion on the at least one first damage detection result to determine the vehicle's damage detection result, including: The single-frame damage detection results and temporal damage detection results are used as input features and input into the trained fusion model. The fusion damage result is generated through the secondary decision model in the fusion model; The secondary decision model is obtained by training and learning the fusion relationship between single-frame damage detection results and temporal damage detection results.

13. The method according to claim 1, characterized in that, After analyzing the vehicle exterior images to determine the damage detection results, the method further includes: A vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results; wherein, the vehicle component detection results include at least one of component type, component location, and component area; By combining the inspection results of the self-propelled vehicle components with the damage inspection results of the vehicle, a vehicle damage assessment result is generated; The vehicle damage assessment results include at least one of the following: damage extent, damage type, damaged components, and repair estimate.

14. The method according to claim 13, characterized in that, The integration of the self-vehicle component inspection results with the vehicle damage inspection results includes: The damaged location in the damage detection results is mapped to the corresponding component location area in the vehicle component detection results to identify the damaged component.

15. The method according to claim 13, characterized in that, Generate vehicle damage assessment results, including: The percentage of the damaged area relative to the area of ​​the corresponding component in the damage detection results is calculated as the degree of damage.

16. The method according to claim 13, characterized in that, Generate vehicle damage assessment results, including: By associating the damage type with the component type in the damage detection results, a damaged component identifier is generated.

17. The method according to claim 16, characterized in that, Generate vehicle damage assessment results, including: The repair estimate is calculated based on the damaged component identification, the degree of damage, and the preset repair knowledge base.

18. The method according to any one of claims 13 to 17, characterized in that, After generating the vehicle damage assessment result, the method further includes: The vehicle damage assessment results are sent to the insurance claims and after-sales service server.

19. The method according to claim 4, characterized in that, In the event of a detected collision signal, the method further includes: A vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results; wherein, the vehicle component detection results include at least one of component type, component location, and component area; The damage discrimination model is used to analyze the detection results of the vehicle parts in order to determine whether the corresponding vehicle parts are damaged.

20. The method according to claim 19, characterized in that, The damage discrimination model is used to analyze the inspection results of the vehicle parts, including: Extract the image features of the vehicle component inspection results; Damage presence is determined based on extracted image features, and the output is a probability value representing the likelihood of damage.

21. The method according to claim 20, characterized in that, Determine whether the corresponding vehicle parts are damaged, including: If the probability value is higher than a set threshold, the corresponding vehicle component is determined to be damaged. If the probability value is not higher than the set threshold, the corresponding vehicle component is determined to be undamaged.

22. The method according to claim 21, characterized in that, In cases where the corresponding vehicle component is determined to be damaged, the method further includes: The appearance image of damaged vehicle parts is detected and segmented using a vehicle damage segmentation network to obtain a second detection result; wherein, the second detection result includes at least one of the damaged part type, the damaged part location, and the damaged part area; By combining the second detection result with the damage detection result of the vehicle, a vehicle damage assessment result is generated; The damage assessment results for the vehicle include at least one of the following: the extent of damage, the type of damage, the damaged components, and the estimated repair cost.

23. The method according to claim 1, characterized in that, Analyzing the vehicle exterior images to determine the vehicle damage detection results includes: A vehicle component detection network is used to detect and segment vehicle exterior images to obtain vehicle component detection results; wherein, the vehicle component detection results include at least one of component type, component location, and component area; The detection network is used to detect and identify the detection results of the vehicle parts to determine the damage detection results of the vehicle. The damage detection results of the vehicle include at least one of the damage type and the damage area.

24. The method according to claim 1, characterized in that, The vision sensor covers at least one of the vehicle's forward field of view, rearward field of view, and lateral field of view.

25. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 24.

26. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 24.

27. A controller having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method described in any one of claims 1 to 24.

28. A vehicle, characterized in that, Includes the controller as described in claim 27, or the steps of performing the method as described in any one of claims 1 to 24.

Images