Vehicle damage degree identification method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a vehicle damage degree identification method and device, equipment and a storage medium, and the method comprises the steps: inputting a to-be-identified vehicle image into an identification model after the to-be-identified vehicle image is obtained, so as to obtain a damage degree score outputted by the identification model; and obtaining a maintenance scheme mapping range, and matching a maintenance scheme in the maintenance scheme mapping range according to the damage degree score to generate maintenance scheme information. According to the method, each maintenance mechanism can adjust the score intervals corresponding to different maintenance schemes according to actual conditions. The damage degree score is used for replacing discrete maintenance scheme output, the damage degree identification result can be refined, the differentiation requirements of vehicle models and maintenance institutions for the maintenance schemes are met, and the accuracy and applicability of the damage assessment result are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a vehicle damage degree identification method, device, equipment and storage medium. Background Art

[0002] Automatic vehicle damage assessment is a damage identification method that combines image processing technology, data analysis and artificial intelligence (AI) algorithms. Automatic vehicle damage assessment can quickly, accurately and comprehensively assess the damage to the vehicle caused by collision, accident or other forms. The results of automatic vehicle damage assessment can be used to determine the repair plan or make insurance claims based on the damage.

[0003] Taking the determination of repair plans as an example, in the process of automatic vehicle damage assessment, the specific damage form can be identified based on the vehicle image, such as scratches, dents, cracks, etc. Then, based on factors such as damage form, damage location, and damage area, a comprehensive repair plan is given, such as painting, small sheet metal, medium sheet metal, large sheet metal, replacement, etc.

[0004] However, in actual business scenarios, due to the influence of factors such as the maintenance level and material prices of different maintenance organizations, the maintenance standards of different organizations are different. For example, the same damage needs to be replaced in one maintenance organization, while in another maintenance organization, a large sheet metal repair method is required. In addition, for the same damage, if the replacement cost is low on a low-end car, a replacement repair plan will be given in the actual damage assessment. On high-end cars, the maintenance cost is high, and no replacement repair plan will be given. Therefore, the damage assessment results obtained by the above-mentioned vehicle automatic damage assessment method cannot accurately adapt to different maintenance organizations and different models, resulting in reduced accuracy and applicability of the damage assessment results. Summary of the invention

[0005] In view of this, the embodiments of the present application provide a method, device, equipment and storage medium for identifying the degree of vehicle damage to solve the problem of reduced accuracy and applicability of the damage assessment results of automatic vehicle damage assessment.

[0006] According to one aspect of the present application, a method for identifying the degree of damage to a vehicle is provided, the method comprising:

[0007] Acquire a vehicle image to be identified;

[0008] Inputting the vehicle image to be identified into a recognition model to obtain a damage degree score output by the recognition model, wherein the recognition model is a neural network model trained based on vehicle damage sample data; the vehicle damage sample data includes a vehicle damage image and a label score marked for the vehicle damage image; the number of output categories of the recognition model is 1, and the recognition model is set to perform pixel-level regression analysis according to a mean square error loss function;

[0009] Acquire a maintenance plan mapping range, wherein the maintenance plan mapping range is a mapping relationship preset according to maintenance capability data of a maintenance organization, the maintenance plan mapping range includes at least one score interval, and the score interval is set with an associated maintenance plan; the maintenance capability data includes at least one of a maintenance level, a material price, and a vehicle model level;

[0010] A maintenance plan is matched within the maintenance plan mapping range according to the damage degree score to generate maintenance plan information.

[0011] According to another aspect of the present application, a vehicle damage degree identification device is provided, the device comprising:

[0012] An image acquisition module, used to acquire the image of the vehicle to be identified;

[0013] A recognition module, used for inputting the vehicle image to be recognized into a recognition model to obtain a damage degree score output by the recognition model, wherein the recognition model is a neural network model trained based on vehicle damage sample data; the vehicle damage sample data includes a vehicle damage image and a label score marked for the vehicle damage image; the number of output categories of the recognition model is 1, and the recognition model is set to perform pixel-level regression analysis according to a mean square error loss function;

[0014] A maintenance scheme mapping module, used for acquiring a maintenance scheme mapping range, wherein the maintenance scheme mapping range is a mapping relationship preset according to maintenance capability data of a maintenance organization, wherein the maintenance scheme mapping range includes at least one score interval, and the score interval is provided with an associated maintenance scheme; the maintenance capability data includes at least one of a maintenance level, a material price, and a vehicle model level;

[0015] A result output module is used to match the maintenance plan in the maintenance plan mapping range according to the damage degree score to generate maintenance plan information.

[0016] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned vehicle damage degree identification method when executing the program.

[0017] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned vehicle damage degree identification method is implemented.

[0018] By means of the above technical solution, the embodiment of the present application provides a method, device, equipment and storage medium for identifying the degree of damage of a vehicle. The method can input the image of the vehicle to be identified into the recognition model after obtaining the image of the vehicle to be identified, so as to obtain the degree of damage score output by the recognition model. Then obtain the maintenance plan mapping range, so as to match the maintenance plan in the maintenance plan mapping range according to the damage degree score to generate maintenance plan information. Among them, the recognition model is a neural network model obtained by training based on vehicle damage sample data, and the recognition model can perform pixel-level regression analysis according to the mean square error loss function. The maintenance plan mapping range is a mapping relationship pre-set according to the maintenance capability data of the maintenance organization. The method can enable each maintenance organization to adjust the score interval corresponding to different maintenance plans according to the actual situation. By replacing the discrete output of several types of maintenance plans with the damage degree score, the damage degree recognition result can be refined to meet the differentiated requirements of vehicle models and maintenance organizations for maintenance plans, and improve the accuracy and applicability of the damage assessment results.

[0019] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0021] Figure 1 A diagram showing an application scenario of vehicle damage degree identification provided by an embodiment of the present application is shown;

[0022] Figure 2 A schematic diagram of a process flow of a vehicle damage degree identification method provided in an embodiment of the present application is shown;

[0023] Figure 3 A schematic diagram of a maintenance plan information generation process provided by an embodiment of the present application is shown;

[0024] Figure 4 A schematic diagram of a damage degree score calculation process provided in an embodiment of the present application is shown;

[0025] Figure 5 Another damage degree score calculation process diagram provided in an embodiment of the present application is shown;

[0026] Figure 6 A schematic diagram of the interval mapping process provided by an embodiment of the present application is shown;

[0027] Figure 7A schematic diagram of the model training process provided in an embodiment of the present application is shown;

[0028] Figure 8 A schematic diagram of a vehicle damage sample data generation process provided by an embodiment of the present application is shown;

[0029] Fig. 9 Another schematic diagram of a vehicle damage sample data generation process provided by an embodiment of the present application is shown;

[0030] Fig.10 A schematic diagram of the structure of a vehicle damage degree identification device provided in an embodiment of the present application is shown;

[0031] Fig.11 A schematic diagram of the structure of a computer device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0032] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0033] In the embodiment of the present application, the vehicle damage degree recognition is an image processing technology. The damage degree of the vehicle can be determined by identifying the features contained in the vehicle image, and then a repair plan can be determined according to the damage degree. The determined repair plan can be applied to application scenarios such as vehicle status assessment, insurance claims, and second-hand car sales.

[0034] In order to identify the degree of damage to the vehicle, in some embodiments, when identifying the degree of damage to the vehicle, the vehicle image can be first obtained through an image acquisition sensor. The vehicle image refers to an image containing the vehicle and the vehicle damage target. The vehicle image can be obtained by taking an image of the damaged vehicle. For example, for an insurance claim application scenario, the vehicle image can be obtained by on-site photography by a car insurance on-site staff using a handheld image acquisition device.

[0035] Depending on the type of vehicle damage, the shooting range of the vehicle image may also be different. In some embodiments, the vehicle image may include the entire vehicle target. For example, for large-area damage on the outer surface of the vehicle, it is necessary to take a photo of the entire vehicle so that the large-area damage area can be completely included in the vehicle image. In addition, in some application scenarios where the vehicle type needs to be identified, it is also necessary to take a photo of the entire vehicle so that the vehicle type recognition model can perform type recognition based on the entire vehicle target.

[0036] In some embodiments, the vehicle image may include a partial area target of the vehicle. For example, for a small area of ​​local damage to the vehicle, the small area of ​​damage may be located in the vehicle image by taking a photo of the local damage area.

[0037] In some embodiments, the vehicle image can be obtained by static image capture, that is, by photographing the vehicle with a camera to obtain static image data. The vehicle image can also be obtained by extracting from a dynamic video, that is, after recording a video of the vehicle with a video capture device, extracting one or more frames from the recorded video data to obtain the vehicle image.

[0038] After acquiring the vehicle image, the vehicle image can be subjected to image recognition. Image recognition is to extract specified features from the vehicle image through an image recognition algorithm. The extracted features are a set of pixels in the vehicle image that have a specific pixel value arrangement pattern. For example, for vehicle surface damage, surface damage features can be extracted from the vehicle image based on the color, texture, shape and other pixel arrangement patterns of the surface damage. When performing feature extraction, different tools can be called for different feature types. For example, shape features are extracted using the Canny and Sobel algorithms; color features are extracted using the Hue-Saturation-Value (HSV) algorithm; texture features are extracted using the Gabor filter and Fourier transform algorithm, etc.

[0039] In some embodiments, feature extraction can be performed by constructing a neural network model. The neural network model can use the convolutional layer in the convolutional neural network (CNN) to extract local features such as edges and textures of the image through filters (or convolution kernels). After the convolutional layer extracts the features, the neural network model can also apply a nonlinear activation function (such as ReLU) to increase the nonlinear expression ability of the model, so that the neural network model can learn more complex features.

[0040] After extracting features from the vehicle image, vehicle damage detection and positioning can be performed based on the extracted features, that is, the damage type, such as scratches, dents, cracks, wrinkles, perforations, etc., can be identified based on the extracted features. And based on the position of the pixel points corresponding to the extracted features in the vehicle image, the position of the damaged area in the entire vehicle can be determined.

[0041] During the vehicle damage detection process, the degree of damage can also be evaluated based on the extracted features. Vehicle damage degree evaluation refers to the process of determining the severity of vehicle damage by performing a comprehensive calculation based on the damage color, texture, shape, etc. corresponding to the extracted features. Among them, the degree of vehicle damage can be associated with the color, texture, and shape of the damage features. The association relationship can be determined based on the correspondence between the specific feature type and the actual damage assessment. For example, the degree of vehicle damage is positively correlated with the damage size in the damage shape, that is, the larger the damage size, the more serious the corresponding degree of vehicle damage.

[0042] In some embodiments, the degree of vehicle damage can be determined in multiple levels according to the set damage assessment standard. For example, the damage assessment standard is determined according to the maintenance plan, and the degree of vehicle damage can be determined in the following five levels according to the maintenance plan. That is, a maintenance plan is given comprehensively based on the damage form, damage location, damage area, etc. The maintenance plan includes painting required, small sheet metal, medium sheet metal, large sheet metal, replacement, etc.

[0043] Among them, painting refers to the repair of minor scratches, paint peeling and other damages on the vehicle by painting; small sheet metal refers to the repair of small dents, scratches or slight deformations on the body surface, such as by knocking, pulling or local sheet metal repair; medium sheet metal refers to the repair of larger dents, scratches or damages that require a certain degree of sheet metal shaping and repair, which requires the use of professional tools and equipment such as beam straighteners and spot welders; large sheet metal refers to the repair of large-area deformation, damage or damage to key parts of the body structure, which requires cutting, welding and other sheet metal work; replacement refers to the repair of structural parts that are difficult to repair by sheet metal repair by replacing structural parts. Obviously, according to the standards of repair difficulty and repair material cost, the higher the degree of damage, the higher the corresponding repair difficulty and repair material cost.

[0044] In order to determine the maintenance plan, a vehicle damage degree recognition model can be constructed. The recognition model is a neural network model obtained by training with training data. The input of the recognition model is a vehicle image, and the output of the recognition model can be set according to specific needs. In some embodiments, the output of the recognition model can be a maintenance plan. For example, the output of the recognition model includes five maintenance plans: painting, small sheet metal, medium sheet metal, large sheet metal, and replacement.

[0045] In the process of identifying the degree of vehicle damage, the vehicle image can be input into the recognition model. The recognition model then performs data preprocessing, feature extraction, feature combination, prediction, evaluation and other classification process calculations based on the internal neural network structure of the model. The classification probability of the vehicle image corresponding to the maintenance plan can be output, and the maintenance plan with the highest classification probability can be used as the vehicle damage degree identification result, that is, the maintenance plan information is generated.

[0046] In some embodiments, the output of the recognition model can also be set according to the damage morphology (type). For example, the output of the recognition model can be set to damage morphology types such as scratches, dents, and cracks. In the process of identifying the degree of vehicle damage, after the vehicle image is input into the recognition model, the recognition model can output the classification probability of the vehicle image corresponding to the damage morphology type, so that the damage morphology type with the highest classification probability is used as the vehicle damage degree recognition result, that is, the maintenance plan information is generated.

[0047] However, in actual business scenarios, due to the influence of factors such as the maintenance level and material prices of different maintenance organizations, the maintenance standards of different maintenance organizations are different. For example, the same damage needs to be replaced in one maintenance organization, while in another maintenance organization, a large sheet metal repair method is required. Moreover, for the same damage, if the replacement cost is low on a low-end car, a replacement repair plan will be given in the actual damage assessment. On high-end cars, the replacement cost is high, and no replacement repair plan will be given. Therefore, the maintenance plan or damage morphology classification result obtained by the vehicle damage degree identification method provided in the above embodiment cannot accurately adapt to different maintenance organizations and different models, resulting in reduced accuracy and applicability of the identification results.

[0048] In order to improve the accuracy and applicability of the recognition results, a vehicle damage degree recognition method is provided in this embodiment, and the vehicle damage degree recognition method can be applied to a vehicle damage degree recognition system. As a software and hardware combination platform for implementing the vehicle damage degree recognition method, the vehicle damage degree recognition system can rely on computers, mobile terminals, servers, industrial hosts, smart wearable devices, etc. to provide users with a human-computer interaction interface, and respond to users to perform collection, processing, transmission, storage and other operations on vehicle images based on the provided human-computer interaction interface to meet the needs of different application fields. That is, the vehicle damage degree recognition system can obtain vehicle images and recognize vehicle images to obtain vehicle damage degree recognition results.

[0049] In a feasible implementation, the vehicle damage degree identification system can be run on a single device, that is, the user can realize all the functions of the vehicle damage degree identification system through a single data processing device. For example, the vehicle damage degree identification system is an application installed on a personal computer. The vehicle damage degree identification system application has a built-in image processing algorithm program. When performing vehicle damage degree identification, the computer can obtain the vehicle image input by the user or collected by the image sensor, and perform image recognition on the vehicle image according to the image processing algorithm program to obtain the corresponding image recognition result data. Then, based on the data processing application in the application field, the image recognition result data is displayed and subsequently processed.

[0050] In another feasible implementation, the vehicle damage degree identification system can be operated by multiple devices, that is, the coordination between multiple devices is required to realize the full functions of the vehicle damage degree identification system. Figure 1 As shown, the vehicle damage degree identification system includes a server and multiple clients connected to the server, and both the client and the server have data processing capabilities. The multiple clients can be operated by different users, and the multiple clients can perform image capture on site, and send the captured vehicle images to the server through the network. The server performs image recognition processing on the vehicle image to obtain an image recognition result. The server then sends the image recognition result data to the client according to the specific application field needs.

[0051] It should be noted that in the embodiment of the present application, a vehicle damage degree identification system is used as the execution subject of the vehicle damage degree identification method. Unless otherwise specified, the vehicle damage degree identification system can be implemented by running relevant applications on the client, or by running relevant applications on the server, or by cooperating with the client to run relevant applications. Therefore, from the perspective of hardware equipment, the steps corresponding to the vehicle damage degree identification method can be executed by the client, or by the server, or some steps can be executed by the client and some steps can be executed by the server.

[0052] like Figure 2 , Figure 3 As shown, the vehicle damage degree identification method includes:

[0053] S101, obtaining a vehicle image to be identified.

[0054] Among them, the vehicle image to be identified is a vehicle image used to identify the degree of vehicle damage. Similar to the vehicle image acquisition method described in the above embodiments, in some embodiments, the vehicle image to be identified can be an image obtained by real-time image acquisition of the damaged vehicle by an image acquisition sensor. For example, when identifying the degree of vehicle damage, an image acquisition request can be sent to a camera set in the damage assessment area. After receiving the image acquisition request, the camera can start image capture to obtain the vehicle image to be identified.

[0055] In some embodiments, the image of the vehicle to be identified can also be obtained by extracting multiple frames of images from a recorded video. For example, multiple cameras for video surveillance can be set up in the vehicle damage assessment area. When identifying the degree of damage to the vehicle, the surveillance video captured in real time by the surveillance camera can be obtained, and the capture time is determined in response to the control instruction for starting the identification of the degree of damage to the vehicle. Thus, the image frames are extracted from the surveillance video according to the capture time to obtain the image of the vehicle to be identified.

[0056] In some embodiments, the image of the vehicle to be identified can also be obtained through user upload. For example, in the insurance industry, after the on-site staff takes a photo of the vehicle at the scene of the accident, the vehicle image obtained by the photo can be uploaded to the insurance service platform. The insurance service platform then sends the vehicle image uploaded by the on-site staff to the client or server according to the insurance claim process when automatic vehicle damage assessment is required, so that the client or server obtains the vehicle image to be identified.

[0057] In order to meet the analysis requirements of the subsequent vehicle damage degree identification process, the vehicle image to be identified should contain at least one type of damage target content, such as scratches, dents, deformations and other damage target content. Therefore, in some embodiments, the vehicle image can be preliminarily processed so that the acquired vehicle image to be identified can meet the content requirements. For example, after obtaining the vehicle images uploaded by the on-site staff, the insurance service platform can screen the uploaded vehicle images, delete the vehicle images with incorrect formats, missing damage targets, and unclear images, and use the vehicle images that meet the preset format, content, and clarity requirements as the vehicle images to be identified.

[0058] In some embodiments, in order to improve the accuracy of the recognition result and reduce the influence of factors such as shooting angle, shooting distance, and shooting position on the recognition result during the image acquisition process, multiple vehicle images to be identified can also be acquired for the same damaged target. The multiple vehicle images to be identified can constitute an acquired associated image set, that is, the associated image set includes multiple vehicle images to be identified for the same damaged target.

[0059] In some embodiments, after obtaining the image of the vehicle to be identified, the image of the vehicle to be identified may also be preprocessed by normalization, data enhancement, and size unification. Among them, normalization can normalize the image data to the same scale, such as [0, 1] or [-1, 1], to speed up the processing speed and improve the generalization ability of the model. Data enhancement processing is to increase the diversity of data by rotation, scaling, cropping, color transformation, etc., to reduce overfitting and improve the robustness of the model. Size unification processing is to adjust images of different sizes to a unified input size to meet the input requirements of the neural network model.

[0060] S102: Input the to-be-recognized vehicle image into a recognition model to obtain a damage degree score output by the recognition model.

[0061] After obtaining the image of the vehicle to be identified, the client can call the recognition model built into the vehicle damage degree recognition system. The recognition model is a neural network model trained based on vehicle damage sample data. In order to meet the requirements of the vehicle damage degree recognition result, the output category number of the recognition model can be set to 1, and the recognition model is set to perform pixel-level regression analysis according to the mean square error loss function.

[0062] For example, based on the semantic segmentation model, the vehicle damage degree recognition function can be improved in a targeted manner to obtain a recognition model. Through targeted improvement, the recognition model can output a score in a specific range. Since the semantic segmentation model can output the category corresponding to each pixel, in the scenario of vehicle damage degree recognition, regression is more suitable than classification because the score is continuous. Therefore, based on semantic segmentation models such as deeplab-v3, the number of categories can be changed from n to 1, and the loss function can be changed from cross entropy loss to mse loss to achieve pixel-level regression.

[0063] The recognition model can be trained and converged by the server, and then called by the client and applied to the vehicle damage degree recognition process. It can also be trained and converged by the client locally and applied to the vehicle loss degree recognition process. It can also be first trained by the server based on public vehicle damage sample data, and then trained by the client based on local personalized vehicle damage sample data, and applied to the vehicle loss degree recognition process.

[0064] After obtaining the recognition model by calling or training, the client or server can obtain the vehicle image to be identified and output the recognition model, so as to perform image recognition on the vehicle image to be identified through the recognition model, so as to obtain the damage degree score corresponding to the image to be identified. Since the label score of the vehicle damage image in the vehicle damage sample data used in the training of the recognition model is within the preset score range, the damage degree score output by the recognition model is also within the preset score range. For example, when the label score of the vehicle damage image in the vehicle damage sample data is a value within the range of [1, 100], the damage degree score output by the recognition model is also a value within the range of [1, 100].

[0065] In some embodiments, in order to perform image content recognition, after the vehicle image to be identified is input into the recognition model, the pre-processing module of the recognition model can perform mask processing on the vehicle image to be identified to determine the area containing damage features in the vehicle image to be identified. Among them, mask processing is the process of converting the vehicle image to be identified into a mask image. The mask image can be a two-dimensional array, and the mask image is the same size as the original image (vehicle image to be identified). The element value of the mask image is used to identify whether the pixel at the corresponding position should participate in a specific processing operation. For example, in a binary mask, a non-zero value (which can be set to 255) represents an area of ​​interest, while a 0 value represents an area of ​​no interest.

[0066] Therefore, after obtaining the image of the vehicle to be identified, the recognition model can generate a mask based on algorithms such as threshold segmentation and edge detection, and use the mask to perform logical operations with the original image, such as AND, OR, NOT, etc., to extract or modify specific areas of the image. For example, in OpenCV, the cv2.bitwise_and() function can be used to apply the mask. If the mask is applied correctly, the area of ​​interest (foreground) will be extracted and the background will be set to black or other specified colors.

[0067] Since the recognition model can perform pixel-level regression on the image of the vehicle to be recognized, that is, for each pixel in the image of the vehicle to be recognized, the corresponding regression value can be calculated, and the pixels in the image of the vehicle to be recognized, including the pixels corresponding to the image outside the vehicle body and the image of the non-damage associated area on the vehicle body, have no effect on the recognition effect of the image. Therefore, the non-background pixels can be determined by identifying the foreground and background colors of the image of the vehicle to be recognized, thereby determining the damage degree score.

[0068] That is Figure 4As shown, in some embodiments, when the image of the vehicle to be identified is input into the recognition model to obtain the damage degree score output by the recognition model, the foreground color and background color can be read from the image of the vehicle to be identified. Then, the non-background pixels in the image of the vehicle to be identified are determined based on the foreground color. Then, the pixel-level regression value of the non-background pixel is calculated based on the recognition model, and the damage degree score is generated based on the pixel-level regression value. Among them, the damage degree score is equal to the average value of the pixel-level regression values ​​corresponding to the non-background pixel points.

[0069] For example, the image recognition model can use OpenCV and other functions to read the image of the vehicle to be identified, that is, read the image of the vehicle to be identified through the cv2.imread() function. Then preprocess the read vehicle image to be identified, including image grayscale, noise reduction, smoothing and other preprocessing operations. That is, use the cv2.cvtColor() function to convert the image into a grayscale image, and use the cv2.GaussianBlur() function to smooth the image. Then, according to the needs of vehicle damage detection, use different methods to extract the foreground and background colors. For example, you can use the threshold segmentation method to set a suitable threshold to segment the image into foreground and background parts, and use the cv2.threshold() function to perform threshold segmentation. Then, according to the characteristics of the foreground and background colors, use the image processing and analysis algorithms provided by OpenCV for identification. For example, use morphological operations, contour detection and other methods to identify the foreground and background colors. According to the recognition results, the foreground and background colors can be marked on the image, that is, use the cv2.drawContours() function to draw contours, and use the cv2.rectangle() function to draw rectangular boxes.

[0070] After determining the foreground color area, the non-background pixels in the vehicle image to be identified can be determined based on the foreground color. For example, the background pixels can be set to a color value of 0 during threshold segmentation, and all pixels with a color value not equal to 0 are determined as non-background pixels. Then, based on the recognition model, pixel-level regression analysis is performed on the non-background pixels to calculate the pixel-level regression values ​​of all non-background pixels, i.e., R PX1 , R PX2 , ..., R PXm Etc., and then calculate the damage degree score, that is, the damage degree score R = (R PX1 +R PX2 +……+R PXm ) / m. Where m is the number of non-background pixels.

[0071] like Figure 5As shown, in some embodiments, when obtaining a set of associated images for the same damaged target, the vehicle images to be identified in the associated image set can be respectively input into the recognition model to obtain multiple damage degree regression values. Then, the average value of the multiple damage degree regression values ​​is calculated to generate the damage degree score.

[0072] For example, when obtaining a vehicle image to be identified, an associated image set for the same damaged target can be obtained, that is, the associated image set can include vehicle image 1 to be identified, vehicle image 2 to be identified, ..., vehicle image N to be identified. Then, the vehicle images to be identified in the associated image set are sequentially input into the recognition model to obtain the damage degree regression value of each vehicle image to be identified, that is, the damage degree regression value R1 of vehicle image 1 to be identified, the damage degree regression value R2 of vehicle image 2 to be identified, ..., the damage degree regression value R N Then calculate the average of the damage degree regression values ​​to obtain the damage degree score, that is:

[0073]

[0074] Among them, R is the damage degree score, N is the number of vehicle images to be identified contained in the associated image set, and R i is the regression value of the damage degree of the i-th vehicle image to be identified.

[0075] It can be seen that in the embodiment of the present application, the damage degree score output by the recognition model is a value within the preset score range, and according to the model accuracy setting, the recognition model can output continuous values ​​within the preset score range. By replacing discrete classification output results with continuous numerical output, the damage degree can be refined, and the model prediction accuracy can be improved while adapting to the different requirements of different maintenance organizations and different vehicle models when determining maintenance plans.

[0076] S103: Obtain maintenance plan mapping range.

[0077] After the recognition model outputs the damage degree score, the maintenance plan mapping range can also be obtained. The maintenance plan mapping range is a mapping relationship pre-set according to the maintenance capability data of the maintenance organization. Different maintenance organizations can pre-set different mapping relationships according to the corresponding maintenance capability data, that is, different maintenance plan mapping ranges can be obtained for different maintenance organizations.

[0078] The maintenance scheme mapping range may include at least one score interval, and the score interval is provided with an associated maintenance scheme. For example, for maintenance organization A, the score intervals and the maintenance schemes associated with each score interval may be preset according to its maintenance capabilities: the score interval of 0-5 corresponds to suspected damage, the score interval of 6-15 corresponds to scratches, the score interval of 16-20 corresponds to scratches and small sheet metal, the score interval of 21-30 corresponds to small sheet metal, the score interval of 31-49 corresponds to medium sheet metal, the score interval of 50-55 corresponds to medium sheet metal and large sheet metal, the score interval of 56-70 corresponds to large sheet metal, the score interval of 71-100 corresponds to replacement, and so on.

[0079] In some embodiments, the maintenance capability data includes at least one of the maintenance level, material price and vehicle model level. That is, when the maintenance organization presets the maintenance plan mapping range, it can comprehensively consider one or more of the maintenance level, material price and vehicle model level. Among them, the maintenance level, material price and vehicle model level can all be set based on the quantitative results of indicators. For example, the maintenance level can be quantified and comprehensively determined based on indicators such as the number, type, model, professional level of maintenance personnel, and years of work of the maintenance equipment owned by the maintenance organization. The material price can be comprehensively set based on factors such as the supply chain situation, transportation and production costs in the area where the maintenance organization is located. The vehicle model level can be comprehensively set based on indicators such as vehicle brand, vehicle price, and vehicle parts supply rate.

[0080] In order to adapt to different actual application situations, different weights can be set for the maintenance capacity data of different items according to factors that may affect the maintenance process, such as time and location. Then, a comprehensive calculation is performed based on the weights to determine the mapping range that better meets the actual situation of the maintenance organization or vehicle model.

[0081] S104: Match a maintenance plan in the maintenance plan mapping range according to the damage degree score to generate maintenance plan information.

[0082] After obtaining the maintenance plan mapping range, the maintenance plan can be matched within the maintenance plan mapping range according to the damage degree score, and the score interval to which the current damage degree score belongs in the maintenance plan mapping range can be determined, so as to determine the corresponding maintenance plan according to the score interval. After determining the maintenance plan, maintenance plan information can be generated based on the determined maintenance plan. The maintenance plan information can be used for display on the client, or it can be directly reflected in the damage assessment result file. For example, the maintenance plan information can be filled in the insurance claim detailed bill to show the current vehicle damage degree and the corresponding maintenance plan.

[0083] like Figure 6As shown, in some embodiments, in the process of matching the maintenance plan in the maintenance plan mapping range according to the damage degree score to generate the maintenance plan information, the damage degree critical value can be first extracted from the maintenance plan mapping range. Then, the damage degree score and the damage degree critical value are compared to determine the first critical value and the second critical value. Among them, the first critical value and the second critical value are the damage degree critical values ​​adjacent to the damage degree score; the first critical value is less than the damage degree score, and the second critical value is greater than the damage degree score.

[0084] For example, after obtaining the maintenance plan mapping range set by maintenance organization B, the damage degree critical value can be read in the maintenance plan mapping range based on the interval score, such as the 0-5 score range corresponds to suspected damage, the 6-15 score range corresponds to abrasions... Then the damage degree critical values ​​can be read as 0, 5, 6, and 15 respectively. After reading the score interval, the difference between the damage degree score and the damage degree critical value can be calculated to determine the first critical value that is less than or equal to the damage degree score and has the smallest difference with the damage degree score. For example, if the damage degree score is 12, the first critical value is calculated to be 6. Similarly, by calculating the difference between the damage degree score and the damage degree critical value, the second critical value that is greater than or equal to the damage degree score and has the smallest difference with the damage degree score can be determined. For example, if the damage degree score is 12, the second critical value is calculated to be 15.

[0085] After determining the first critical value and the second critical value, the target score interval can be matched according to the first critical value and the second critical value, and the maintenance plan associated with the target score interval can be obtained, and the maintenance plan information can be generated according to the maintenance plan. For example, after determining that the first critical value is 6 and the second critical value is 15, the target score interval is determined to be 6-15, so the maintenance plan corresponding to the score interval can be scratch paint repair, that is, the maintenance plan information containing "scratched paint" is generated.

[0086] By applying the technical solutions of the above embodiments, the vehicle damage degree recognition method provided in the above embodiments can input the vehicle image to be recognized into the recognition model after obtaining the vehicle image to be recognized, so as to obtain the damage degree score output by the recognition model. Then obtain the maintenance plan mapping range, so as to match the maintenance plan in the maintenance plan mapping range according to the damage degree score to generate maintenance plan information. Among them, the recognition model is a neural network model obtained by training based on vehicle damage sample data, and the recognition model can perform pixel-level regression analysis according to the mean square error loss function. The maintenance plan mapping range is a mapping relationship pre-set according to the maintenance capability data of the maintenance organization. The method can enable each maintenance organization to adjust the score interval corresponding to different maintenance plans according to the actual situation. By replacing the discrete output of several types of maintenance plans with the damage degree score, the damage degree recognition result can be refined to meet the differentiated requirements of vehicle models and maintenance organizations for maintenance plans, and improve the accuracy and applicability of damage assessment results.

[0087] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, some embodiments of this application also provide a recognition model training method, which can perform model training before inputting the vehicle image to be identified into the recognition model to obtain a recognition model with a certain output result accuracy. Figure 7 As shown, the method includes:

[0088] S201, obtaining the vehicle damage sample data;

[0089] When executing model training, the client or server for executing model training can first obtain the vehicle damage sample data. The vehicle damage sample data includes a vehicle damage image and a label score marked for the vehicle damage image. For example, in order to refine the degree of damage to a score in the interval [1, 100] so that the final recognition model can output a damage score, the vehicle damage image A in the vehicle damage sample data can be first marked as a score in the interval [1, 100] according to the degree of damage, such as 25, through sample annotation. In this way, the vehicle damage image A and the label score 25 marked for the vehicle damage image A are obtained. Based on this, a large number of vehicle damage images can be annotated to form vehicle damage sample data with a certain data scale.

[0090] like Figure 8As shown, in some embodiments, in order to obtain vehicle damage sample data, the sample data may be annotated, that is, a vehicle damage image may be first obtained, an image mask may be generated based on the vehicle damage image, and a damage area may be delineated from the vehicle damage image based on the image mask. Damage parameters may then be extracted based on the damage area, wherein the damage parameters include at least one of a damage morphology parameter, a damage type parameter, a damage location parameter, and an empirical value. The label score may then be calculated based on the damage parameter, and the label score may be annotated to the vehicle damage image to generate the vehicle damage sample data.

[0091] That is, when training the model, a vehicle damage image can be obtained, and the damaged area can be determined based on the image mask processing, and then the damage parameters can be calculated based on the pixel color value, texture, edge and other features corresponding to the damaged target in the damaged area. The damage parameters can include damage morphological parameters, such as the shape of the damaged area, the length of the damaged area, the area of ​​the damaged area, etc. The damage parameters can also include damage type parameters, that is, the damage type is determined by detecting the pixel color, edge, texture and other features of the damaged area, including abrasions, scratches, bruises, etc. These damage types can be quantified into specific numerical representations to facilitate label score calculation. The damage parameters can also include damage location parameters, that is, the location of the damaged area in the image. Damage evaluation can also be performed based on certain empirical values, that is, the damage parameters can also include empirical values.

[0092] After extracting the damage parameters, the label score can be calculated based on the damage parameters, that is, the weighted sum of each item in the damage parameters is calculated to obtain the label score, and it is annotated to the vehicle damage image to generate vehicle damage sample data.

[0093] like Fig. 9 As shown, in some embodiments, the label score can also be annotated by the maintenance agency, that is, the vehicle damage image can be scored according to multiple maintenance agencies, and the final label score can be calculated according to the multiple damage score results, so as to obtain the vehicle damage sample data. When generating the vehicle damage sample data, the vehicle damage image and the damage scores sent by multiple maintenance agencies for the vehicle damage image can be obtained first. Then the maintenance capability data of the maintenance agency is obtained, and the score weight of the maintenance agency is calculated according to the maintenance capability data. The label score is calculated according to the score weight and the damage score, and the label score is annotated to the vehicle damage image to generate the vehicle damage sample data. Among them, the label score is the weighted average of the multiple damage scores based on the score weight.

[0094] For example, for a vehicle damage image C, the data processing device that performs model training can send the vehicle damage image C to a maintenance organization, and the maintenance personnel or the evaluation system of the maintenance organization can perform damage scoring, that is, obtain multiple damage scores, P1, P2, ..., P nThen obtain the maintenance capability data of multiple maintenance organizations, and calculate the scoring weights of the maintenance organizations according to the maintenance capability data, i.e., W1, W2, ..., W n Finally, by calculating the weighted average, we can get the label score P = (W1*P1+W2*P2+…W n *P n ) / n.

[0095] S202: Input the vehicle damage sample data into a trained model to obtain a training score output by the trained model.

[0096] After obtaining the vehicle damage sample data, the vehicle damage sample data can be input into the trained model to obtain the training score output by the trained model. The neural network structure contained in the trained model is the same as that of the recognition model, that is, the trained model can be an initial neural network model or an unconverged neural network model with the same input and output structure as the recognition model. Similar to the recognition model, the number of output categories of the trained model is set to 1, and the loss function of the trained model is set to the mean square error loss function. That is, the trained model can also change the number of output categories from n to 1 and the loss function from cross entropy loss to mse loss based on semantic segmentation models such as deeplab-v3 to achieve pixel-level regression.

[0097] After the vehicle damage sample data is input into the trained model, the trained model can perform pixel-level regression analysis on the vehicle damage sample data to obtain a training score. The training score is also a single score in a preset interval such as [1, 100], which is used to represent the vehicle damage score.

[0098] S203: Calculate training loss according to the training score and the label score.

[0099] After obtaining the training score, the mean square error (MSE) can be calculated based on the training score and the label score to obtain the training loss. The training loss is the mean square error calculated based on the training score and the label score, that is, the training loss can be calculated as follows:

[0100]

[0101] Where n is the number of samples; y i is the actual value of the i-th sample, that is, the label score; is the predicted value of the i-th sample, that is, the training score.

[0102] S204: If the training loss is greater than a loss threshold, modify the model parameters of the trained model according to the training loss.

[0103] After the training loss is calculated, the training loss can be compared with a preset loss threshold. If the training loss is greater than the loss threshold, it means that the current trained model has not converged, and the model parameters of the trained model can be modified according to the training loss, and iterative training can be continued based on the modified model parameters until the training loss is less than or equal to the loss threshold.

[0104] S205. If the training loss is less than or equal to the loss threshold, output the model parameters of the trained loss to generate the recognition model.

[0105] By comparing the training loss and the loss threshold, if the training loss is less than or equal to the loss threshold, it means that the current trained model has converged, so the model parameters of the trained loss can be output to generate the recognition model.

[0106] It should be noted that, in some embodiments, when performing model training, a threshold of the number of iterations can be preset according to the actual accuracy requirements. When the number of iterations of the iterative training reaches the preset threshold of the number of iterations, the model parameters of the trained loss can also be output to generate the recognition model.

[0107] After obtaining the recognition model based on the model training method provided in the above embodiment, the vehicle image to be recognized can be obtained by referring to the method described in the embodiment corresponding to the above vehicle damage degree recognition method, and the vehicle image to be recognized is input into the recognition model to obtain the damage degree score output by the recognition model. Then, the maintenance plan mapping range is obtained, and the maintenance plan mapping range is matched according to the damage degree score to generate maintenance plan information.

[0108] Further, as a specific implementation of the vehicle damage degree identification method described in the above embodiment, the present application embodiment provides a vehicle damage degree identification device, such as Fig.10 As shown, the device comprises:

[0109] An image acquisition module, used to acquire the image of the vehicle to be identified;

[0110] A recognition module, used for inputting the vehicle image to be recognized into a recognition model to obtain a damage degree score output by the recognition model, wherein the recognition model is a neural network model trained based on vehicle damage sample data; the vehicle damage sample data includes a vehicle damage image and a label score marked for the vehicle damage image; the number of output categories of the recognition model is 1, and the recognition model is set to perform pixel-level regression analysis according to a mean square error loss function;

[0111] A maintenance scheme mapping module, used for acquiring a maintenance scheme mapping range, wherein the maintenance scheme mapping range is a mapping relationship preset according to maintenance capability data of a maintenance organization, wherein the maintenance scheme mapping range includes at least one score interval, and the score interval is provided with an associated maintenance scheme; the maintenance capability data includes at least one of a maintenance level, a material price, and a vehicle model level;

[0112] A result output module is used to match the maintenance plan in the maintenance plan mapping range according to the damage degree score to generate maintenance plan information.

[0113] It should be noted that for other corresponding descriptions of the functional units involved in the vehicle damage degree identification device provided in the embodiment of the present application, reference can be made to the corresponding descriptions in the vehicle damage degree identification method provided in the above embodiment, and will not be repeated here.

[0114] like Fig.11 As shown, the embodiment of the present application also provides a computer device, which can be specifically a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and can also include an input and output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0115] Those skilled in the art will appreciate that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0116] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0117] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0118] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0119] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0120] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for identifying the degree of vehicle damage, characterized in that: The method comprises: Obtaining a vehicle image to be identified; Inputting the vehicle image to be identified into a recognition model to obtain a damage degree score output by the recognition model, wherein the recognition model is a neural network model trained based on vehicle damage sample data; the vehicle damage sample data includes a vehicle damage image and a label score marked for the vehicle damage image; the number of output categories of the recognition model is 1, and the recognition model is set to perform pixel-level regression analysis according to a mean square error loss function; Acquire a maintenance plan mapping range, wherein the maintenance plan mapping range is a mapping relationship preset according to maintenance capability data of a maintenance organization, the maintenance plan mapping range includes at least one score interval, and the score interval is set with an associated maintenance plan; the maintenance capability data includes at least one of a maintenance level, a material price, and a vehicle model level; A maintenance plan is matched within the maintenance plan mapping range according to the damage degree score to generate maintenance plan information.

2. The method according to claim 1, characterized in that The method further comprises: Acquiring the vehicle damage sample data; Input the vehicle damage sample data into a trained model to obtain a training score output by the trained model, wherein the neural network structure included in the trained model is the same as that of the recognition model; and the number of output categories of the trained model is set to 1; Calculate a training loss based on the training score and the label score, where the training loss is a mean square error calculated based on the training score and the label score; If the training loss is greater than a loss threshold, modifying a model parameter of the trained model according to the training loss; If the training loss is less than or equal to the loss threshold, the model parameters of the trained loss are output to generate the recognition model.

3. The method according to claim 2, characterized in that Acquiring the vehicle damage sample data, including: Obtain vehicle damage images; generating an image mask according to the vehicle damage image; Delineating a damaged area from the vehicle damage image based on the image mask; Extracting damage parameters according to the damage area, wherein the damage parameters include at least one of a damage morphology parameter, a damage type parameter, a damage location parameter, and an empirical value; The label score is calculated according to the damage parameter, and the label score is annotated to the vehicle damage image to generate the vehicle damage sample data.

4. The method according to claim 2, characterized in that: Acquiring the vehicle damage sample data, including: Acquire a vehicle damage image and damage scores sent by multiple maintenance agencies for the vehicle damage image; Acquiring maintenance capability data of the maintenance organization, and calculating a scoring weight of the maintenance organization according to the maintenance capability data; Calculate the label score according to the score weight and the damage score, wherein the label score is a weighted average of multiple damage scores based on the score weight; The label score is annotated to the vehicle damage image to generate the vehicle damage sample data.

5. The method according to claim 1, characterized in that Inputting the image of the vehicle to be identified into a recognition model to obtain a damage degree score output by the recognition model includes: Reading the foreground color and the background color from the image of the vehicle to be identified; Determine non-background pixel points in the image of the vehicle to be identified according to the foreground color; Calculate the pixel-level regression value of the non-background pixel based on the recognition model; The damage degree score is generated according to the pixel-level regression value, and the damage degree score is equal to the average value of the pixel-level regression values ​​corresponding to the non-background pixel points.

6. The method according to claim 1, characterized in that Inputting the to-be-recognized vehicle image into a recognition model to obtain a damage degree score output by the recognition model, the method further comprising: Acquire an associated image set, wherein the associated image set includes a plurality of vehicle images to be identified for a same damaged target; Inputting the to-be-recognized vehicle images in the associated image set into the recognition model respectively to obtain a plurality of damage degree regression values; An average of a plurality of the damage degree regression values ​​is calculated to generate the damage degree score.

7. The method according to claim 1, characterized in that Matching a maintenance plan in the maintenance plan mapping range according to the damage degree score to generate maintenance plan information, including: Extracting a damage degree critical value from the maintenance solution mapping range; Comparing the damage degree score with the damage degree critical value to determine a first critical value and a second critical value; the first critical value and the second critical value are the damage degree critical values ​​adjacent to the damage degree score; the first critical value is less than the damage degree score, and the second critical value is greater than the damage degree score; Matching a target score interval according to the first critical value and the second critical value; A maintenance plan associated with the target score interval is obtained, and the maintenance plan information is generated according to the maintenance plan.

8. A vehicle damage degree identification device, characterized in that: The device comprises: An image acquisition module, used to acquire the image of the vehicle to be identified; A recognition module, used for inputting the vehicle image to be recognized into a recognition model to obtain a damage degree score output by the recognition model, wherein the recognition model is a neural network model trained based on vehicle damage sample data; the vehicle damage sample data includes a vehicle damage image and a label score marked for the vehicle damage image; the number of output categories of the recognition model is 1, and the recognition model is set to perform pixel-level regression analysis according to a mean square error loss function; A maintenance scheme mapping module, used for acquiring a maintenance scheme mapping range, wherein the maintenance scheme mapping range is a mapping relationship preset according to maintenance capability data of a maintenance organization, wherein the maintenance scheme mapping range includes at least one score interval, and the score interval is provided with an associated maintenance scheme; the maintenance capability data includes at least one of a maintenance level, a material price, and a vehicle model level; A result output module is used to match the maintenance plan in the maintenance plan mapping range according to the damage degree score to generate maintenance plan information.

9. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images