Intelligent vehicle loss measurement identification data processing method and system

Abstract

The invention discloses an intelligent vehicle loss measurement identification data processing method and system. The method relates to the technical field of vehicle loss detection identification data processing, and comprises the following steps: quantifying detection precision; and target detection precision optimization, quantization edge precision and segmentation stage edge precision optimization are carried out. According to the method, the vehicle image is acquired and the target is detected; the vehicle damage identification accuracy is evaluated according to the detection precision parameter; whether the target detection precision is optimized or not is judged; the damage area segmentation is performed after optimization or is directly performed, and then the real damage edge adaptation precision is evaluated and predicted according to the segmentation edge precision parameter, and whether the segmentation edge precision is optimized or not is judged; the vehicle damage identification result is output directly or after optimization, the processing precision of the vehicle damage detection identification data is improved, and the problem that the processing precision of the vehicle damage detection identification data is low due to inaccurate target detection in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of vehicle damage detection and identification data processing technology, and in particular to an intelligent vehicle damage detection and identification data processing method and system. Background Technology

[0002] First, vehicle damage images are collected through multiple channels (such as historical claims data, partner repair shops, and targeted photography). Then, rigorous data preprocessing and cleaning are performed, including deduplication, format standardization, image enhancement (such as adjusting brightness and contrast), and size normalization, to ensure data quality. Next comes the crucial data annotation stage, where a professional annotation platform such as LabelMe is used. Annotation methods, such as bounding box annotation, polygon annotation, and pixel-level mask annotation, are selected based on the damage type. The annotated images are then randomly transformed to generate new samples that retain damage characteristics.

[0003] The processed dataset was then divided into training, validation, and test sets according to a set ratio, using deep learning-based object detection and instance segmentation models. The object detection model outputs bounding boxes of damaged areas. Single-stage detection models like the YOLO series (such as YOLOv5 and YOLOv8) can quickly identify vehicle damage in real-time scenarios such as mobile apps due to their fast inference speed. They can also efficiently detect multiple damaged areas in a single image, adapting well to both regular and irregular shapes, providing support for rapid preliminary damage assessment at the scene. Faster R-CNN, a two-stage detection model, generates candidate regions through a region proposal network and performs fine-tuning, resulting in higher bounding box localization accuracy. This plays a crucial role in scenarios such as insurance claims where precise determination of the damage range is needed to calculate repair costs, ensuring the accuracy of damage assessment. The instance segmentation model can output pixel-level masks. Mask R-CNN adds a mask branch on the basis of Faster R-CNN, which can not only output bounding boxes, but also generate pixel-level masks to accurately distinguish between damaged and normal areas. For minor damage such as paint swirl and glass cracks, as well as irregular damage such as fender wrinkles, it can clearly outline the contours, providing detailed basis for subsequent detailed assessment of the degree of damage and development of repair plans.

[0004] For example, the Chinese invention patent with announcement number CN113780435B discloses a vehicle damage detection method, device, equipment, and storage medium, which includes: acquiring a damaged vehicle image, inputting the damaged vehicle image into a trained integrated damage assessment model to obtain damaged component information, obtaining the damage location corresponding to the damaged vehicle image according to the vehicle component segmentation model, and obtaining the vehicle damage information according to the damage location and damaged component information.

[0005] For example, Chinese invention patent CN115099097B discloses a perception-based vehicle damage prediction method, system, electronic device, and vehicle, including: S1 determining the collision direction, S2 determining the collision location, S3 predicting the deformation of key points, and S4 identifying damaged components. The system obtains relevant information before and after the collision from the perception module, the vehicle information module, and the acceleration sensor module. It calculates the vehicle's deformation area and the severity of the accident through the onboard computing module, and determines a list of damaged components based on the component coordinate information stored in the memory module, ultimately achieving the prediction of the vehicle's collision direction, severity, and damaged components.

[0006] The above-mentioned technology has at least the following technical problems: In the vehicle damage detection and localization process, during the target detection stage, the model may fail to identify real damage due to weak damage features, complex background, or poor lighting, resulting in missed detection. Alternatively, it may misidentify non-damaged areas (such as reflections, stains, or vehicle logos) as damage, leading to false detection. Due to the scarcity of high-quality labeled samples, especially the lack of fine labeling for small and blurry damage, the anchor box mechanism and non-maximum suppression processing relied upon by the target detection task perform poorly on dense or small target damage. This directly leads to the inability to perform subsequent instance segmentation or the segmentation of incorrect targets.

[0007] During the instance segmentation stage, even if the target detection is correct, the model often results in rough edges, excessive expansion or contraction of the segmented damage area due to blurred damage edges, low contrast with the background, or inaccurate boundaries in the labeled data. This makes it impossible to accurately fit the real damage contour. The instance segmentation model loses detailed information through multiple downsampling during the feature extraction process, which further exacerbates the difficulty of feature extraction and boundary determination. There is a problem of low accuracy in vehicle damage identification data processing due to inaccurate target detection. Summary of the Invention

[0008] To address the problem of low accuracy in vehicle damage detection data processing due to inaccurate target detection in existing technologies, this invention provides an intelligent vehicle damage detection data processing method and system. The technical solution is as follows: On the one hand, an intelligent vehicle damage detection and identification data processing method is provided, including the following steps: acquiring vehicle image data, performing target detection on the vehicle image to obtain detection accuracy parameters, obtaining target detection accuracy values ​​based on the detection accuracy parameters to evaluate the accuracy of identifying vehicle damage during the target detection process; determining whether to perform target detection accuracy optimization based on the target detection accuracy value, if yes, then performing a vehicle damage region segmentation process after target detection accuracy optimization, if no, directly performing a vehicle damage region segmentation process, and obtaining edge accuracy parameters in the vehicle damage region segmentation process, obtaining edge accuracy values ​​at the segmentation stage based on the edge accuracy parameters to evaluate the matching accuracy between the predicted vehicle damage edge and the actual vehicle damage edge, the target detection accuracy optimization including dynamic adjustment of the bounding box rotation angle and dynamic adjustment of the direction error recovery time reference interval; determining whether to perform segmentation stage edge accuracy optimization based on the segmentation stage edge accuracy value, if yes, then outputting the vehicle damage identification result after segmentation stage edge accuracy optimization, if no, directly outputting the vehicle damage identification result, the segmentation stage edge accuracy optimization including dynamic adjustment of the boundary symmetry distance and dynamic adjustment of the edge root mean square error qualification threshold.

[0009] On the other hand, an intelligent vehicle damage detection and identification data processing system is provided, including: a detection accuracy evaluation module, a target detection accuracy optimization and edge accuracy evaluation module, and a segmentation stage edge accuracy optimization module. The detection accuracy evaluation module acquires vehicle image data and performs target detection on the vehicle images to obtain detection accuracy parameters. Based on the detection accuracy parameters, it obtains a target detection accuracy value to evaluate the accuracy of identifying vehicle damage during the target detection process. The target detection accuracy optimization and edge accuracy evaluation module determines whether to perform target detection accuracy optimization based on the target detection accuracy value. If so, it performs vehicle damage region segmentation after target detection accuracy optimization; otherwise, it directly performs vehicle damage region segmentation and acquires edge accuracy parameters in the vehicle damage region segmentation process. Based on the edge accuracy parameters, it obtains a segmentation stage edge accuracy value to evaluate the matching accuracy between the predicted vehicle damage edge and the actual vehicle damage edge. The segmentation stage edge accuracy optimization module determines whether to perform segmentation stage edge accuracy optimization based on the segmentation stage edge accuracy value. If so, it outputs the vehicle damage identification result after segmentation stage edge accuracy optimization; otherwise, it directly outputs the vehicle damage identification result.

[0010] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. Acquire vehicle image data and perform target detection on the vehicle images to obtain detection accuracy parameters. This accurately assesses the accuracy of vehicle damage recognition and provides data support for subsequent detection accuracy optimization. The target detection accuracy value obtained based on the detection accuracy parameters is used to evaluate the accuracy of vehicle damage recognition during the target detection process. The system determines whether to perform target detection accuracy optimization based on the target detection accuracy value. If so, vehicle damage region segmentation is performed after target detection accuracy optimization; otherwise, vehicle damage region segmentation is performed directly to ensure optimal recognition accuracy and improve the overall recognition process accuracy. The system also obtains edge accuracy values ​​based on edge accuracy parameters to evaluate the matching accuracy between predicted vehicle damage edges and actual vehicle damage edges. This accurately reflects the quality of the segmentation results and provides a quantitative basis for subsequent edge optimization. The system determines whether to perform edge accuracy optimization based on the segmentation stage edge accuracy value. If so, vehicle damage recognition results are output after segmentation stage edge accuracy optimization; otherwise, vehicle damage recognition results are output directly to ensure that the final output vehicle damage recognition results have high-precision edge matching, improving the reliability and practicality of the recognition results, and thus improving the accuracy of vehicle damage detection and recognition data processing.

[0011] 2. The system determines whether to perform target detection accuracy optimization based on the target detection accuracy value. This allows for dynamic adjustment of the bounding box rotation angle and direction error recovery time during the detection process, effectively improving the accuracy and robustness of vehicle damage target recognition, reducing false detections and missed detections, and providing a more reliable input basis for subsequent damage region segmentation. It also determines whether to perform dynamic adjustment of the bounding box rotation angle based on the false detection density per unit image area, effectively reducing the false detection rate, improving the positioning accuracy of target detection, reducing misidentifications caused by angle deviations, and enhancing the stability and reliability of detection results, providing a more accurate target region input for subsequent damage region segmentation. Finally, it determines whether to perform dynamic adjustment of the direction error recovery time reference interval based on the direction error accumulation rate, effectively mitigating the performance degradation caused by the continuous accumulation of direction errors, preventing error propagation over time, thereby improving the directional stability and robustness of target detection, providing a more reliable directional basis for subsequent damage recognition, reducing false positives and missed detections, and ultimately improving the accuracy of vehicle damage detection data processing.

[0012] 3. Determining whether to perform edge precision optimization based on the edge precision value during the segmentation stage can effectively improve the accuracy and robustness of vehicle damage area segmentation; determining whether to perform dynamic adjustment of boundary symmetry distance based on boundary offset error can effectively improve the positioning accuracy of segmentation edges in vehicle damage areas, gradually reduce edge deviation, make the segmentation edges more closely match the actual damage contour, and reduce misjudgments and omissions; determining whether to perform dynamic adjustment of the edge root mean square error qualification threshold based on the average deviation distance of edge pixels can achieve adaptive optimization of segmentation edge quality to adapt to the segmentation difficulty and edge complexity of the current image, ensuring that high edge matching accuracy is maintained under different scenarios (such as changes in illumination, background interference, and irregular damage shapes), thereby improving the accuracy of vehicle damage detection and identification data processing. Attached Figure Description

[0013] Figure 1 A flowchart illustrating an intelligent vehicle damage detection and identification data processing method provided in this application embodiment; Figure 2 A flowchart illustrating the dynamic adjustment of the direction error recovery time reference interval for an intelligent vehicle damage detection and identification data processing method provided in this application embodiment; Figure 3 A flowchart illustrating the dynamic adjustment of boundary symmetric distance in an intelligent vehicle damage detection and identification data processing method provided in this application embodiment; Figure 4 This is a schematic diagram of the structure of an intelligent vehicle damage detection and identification data processing system provided in an embodiment of this application. Detailed Implementation

[0014] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0015] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0016] This application provides an intelligent vehicle damage detection and identification data processing method and system, which solves the problem of low accuracy in vehicle damage detection and identification data processing due to inaccurate target detection in the prior art. By acquiring vehicle images and performing target detection, the accuracy of vehicle damage identification is evaluated based on the detection accuracy parameters. It is then determined whether to optimize the target detection accuracy. After optimization, or by directly segmenting the damage area, the accuracy of the predicted and actual damage edges is evaluated based on the segmentation edge accuracy parameters. It is then determined whether to optimize the segmentation edge accuracy. After optimization, or by directly outputting the vehicle damage identification result, the accuracy of vehicle damage detection and identification data processing is improved.

[0017] The technical solution in this application embodiment aims to address the aforementioned problem of low accuracy in vehicle damage identification data processing due to inaccurate target detection. The overall approach is as follows: By acquiring vehicle image data and performing target detection on the vehicle images to obtain detection accuracy parameters, the target detection accuracy value is obtained based on the detection accuracy parameters to evaluate the accuracy of vehicle damage identification during the target detection process. Based on the target detection accuracy value, it is determined whether to perform target detection accuracy optimization. If yes, the vehicle damage region segmentation process is performed after target detection accuracy optimization; otherwise, the vehicle damage region segmentation process is performed directly, and the edge accuracy parameters in the vehicle damage region segmentation process are obtained. Based on the edge accuracy parameters, the edge accuracy value of the segmentation stage is obtained to evaluate the matching accuracy between the predicted vehicle damage edge and the actual vehicle damage edge. Based on the edge accuracy value of the segmentation stage, it is determined whether to perform segmentation stage edge accuracy optimization. If yes, the vehicle damage identification result is output after segmentation stage edge accuracy optimization; otherwise, the vehicle damage identification result is output directly. Segmentation stage edge accuracy optimization includes dynamic adjustment of boundary symmetry distance and dynamic adjustment of the edge root mean square error qualification threshold, improving the accuracy of vehicle damage identification data processing.

[0018] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0019] This invention provides an intelligent vehicle damage detection and identification data processing method, such as... Figure 1 The flowchart shown is a method for intelligent vehicle damage detection and identification data processing. The flowchart includes the following steps: The first step of this intelligent vehicle damage detection and identification data processing method is to acquire vehicle image data, perform target detection on the vehicle image to obtain detection accuracy parameters, and obtain target detection accuracy values ​​based on the detection accuracy parameters to evaluate the accuracy of identifying vehicle damage during the target detection process.

[0020] It needs to be explained that the specific steps to obtain the target detection accuracy value include: The detection accuracy parameters include bounding box offset, damage range error rate, and minimum bounding box distance. Specifically, the bounding box offset is obtained by calculating the Euclidean distance between the predicted bounding box of the damaged area in the image and the actual bounding box coordinates of the corresponding damaged area in the labeled data; the damage range error rate is obtained by calculating the ratio of the difference between the set value of the vehicle damage area and the actual vehicle damage area to the actual area; and the minimum bounding box distance is obtained by calculating the shortest pixel distance between the bounding box and the actual damage boundary.

[0021] The bounding box offset impact value is obtained by combining the results of the relative deviation ratio between the bounding box offset and the set bounding box offset value using the bounding box offset compensation factor. The relative deviation ratio between the bounding box offset and the set bounding box offset value refers to the ratio of the absolute value of the difference between the bounding box offset and the set bounding box offset value to the set bounding box offset value.

[0022] The error rate impact value is obtained by combining the results of the relative deviation ratio between the damage range error rate and the damage range error rate set value using the error rate compensation factor. The result of the relative deviation ratio between the damage range error rate and the damage range error rate set value refers to the ratio of the absolute value of the difference between the damage range error rate and the damage range error rate set value to the damage range error rate set value.

[0023] The minimum distance influence value is obtained by combining the results of the minimum distance of the bounding box and the ratio of the minimum distance setpoint of the bounding box with the minimum distance compensation factor. Here, the ratio processing is represented by a division operation.

[0024] The target detection accuracy value is obtained by coupling the bounding box offset influence value, the error rate influence value, and the minimum distance influence value. Here, combination operation represents multiplication, and coupling represents addition.

[0025] It's important to understand that the bounding box offset compensation factor, bounding box offset setting, error rate compensation factor, damage range error rate setting, minimum distance compensation factor, and minimum bounding box distance setting are all obtained from the damage detection and identification database. Among these, the detection accuracy parameters are correlated as follows: the bounding box offset reflects the degree of deviation between the predicted bounding box and the actual target in spatial position, directly affecting the accuracy of target localization; the damage range error rate measures the matching error between the predicted damage area and the actual damage area in terms of area or coverage, reflecting the accuracy of damage range identification; and the minimum bounding box distance is used to evaluate the shortest spatial distance between the predicted box and the actual box, and is an important indicator of the tightness of target localization. All three work together; a large bounding box offset or a large minimum bounding box distance often leads to an increased damage range error rate, indicating inaccurate target localization, which in turn affects the reliability of subsequent damage segmentation and identification. The detection accuracy parameters are correlated with the target detection accuracy value, as follows: the larger the bounding box offset, the more serious the deviation between the predicted box and the real target position, and the higher the target detection accuracy value, showing a positive correlation; the higher the damage range error rate, the worse the matching degree between the predicted damage area and the actual damage area, and the higher the target detection accuracy value, also showing a positive correlation; while the smaller the minimum distance of the bounding box, the closer the predicted box and the real box are in space, and the lower the target detection accuracy value, showing a negative correlation.

[0026] In this embodiment, by comprehensively analyzing three detection accuracy parameters—bounding box offset, damage range error rate, and minimum bounding box distance—and combining and coupling them with their respective compensation factors and set values, a target detection accuracy value for quantitative evaluation of target detection accuracy is constructed. The technical effects are: it not only achieves a multi-dimensional quantitative evaluation of the spatial positioning accuracy and damage range recognition precision during vehicle damage target detection, but also improves the model's adaptability to different damage morphologies and image conditions through a dynamic compensation mechanism; simultaneously, this evaluation mechanism effectively reflects the degree of deviation between the target detection results and the actual damage, providing a scientific basis for subsequent adjustment strategies such as bounding box rotation angle recovery and edge accuracy optimization, thereby significantly improving the accuracy of the vehicle damage recognition system.

[0027] The second step of this intelligent vehicle damage detection and identification data processing method is as follows: Based on the target detection accuracy value, it is determined whether to perform target detection accuracy optimization. If so, the vehicle damage area segmentation process is performed after target detection accuracy optimization. If not, the vehicle damage area segmentation process is performed directly, and the edge accuracy parameters in the vehicle damage area segmentation process are obtained. The edge accuracy value of the segmentation stage is obtained based on the edge accuracy parameters to evaluate the matching accuracy between the predicted vehicle damage edge and the real vehicle damage edge. Target detection accuracy optimization includes dynamic adjustment of the bounding box rotation angle and dynamic adjustment of the direction error recovery time reference interval.

[0028] The specific steps for obtaining the edge precision value during the segmentation stage are as follows: Edge accuracy parameters include target detection accuracy, edge intersection-union ratio (IU), and edge pixel accuracy. The specific steps for obtaining these parameters are as follows: The IU is obtained by extracting edge pixels from the segmentation results and comparing them with the actual damaged edge pixels in the ground truth annotations, calculating the ratio of their intersection to their union; and by calculating the ratio of pixels correctly identified as real edges in the segmented edges to the total number of pixels.

[0029] The detection accuracy impact value is obtained by combining the results of the target detection accuracy setpoint and the target detection accuracy value ratio processing by the detection accuracy compensation factor. Here, the ratio processing represents a division operation.

[0030] The cross-union ratio (CUNR) influence value is obtained by combining the results of the edge CUNR and edge CUNR setpoint ratio processing with the cross-union ratio compensation factor. Here, the ratio processing represents a division operation.

[0031] The accuracy impact value is obtained by combining the results of the accuracy compensation factor processing of edge pixel accuracy and the ratio processing of edge pixel accuracy setpoint. Here, the ratio processing is represented by a division operation.

[0032] The edge accuracy value of the segmentation stage is obtained by coupling the impact values ​​of detection accuracy, intersection-over-union ratio, and accuracy. Here, combination operation represents multiplication, and coupling represents addition.

[0033] It should be noted that the detection accuracy compensation factor, edge intersection-union ratio (IU / U), target detection accuracy setting, accuracy compensation factor, and edge pixel accuracy setting are all obtained from the damage detection and identification database. Among these, the edge accuracy parameters are correlated, specifically as follows: Target detection accuracy is usually measured by the IU / U, which is the ratio of the intersection to the union of the predicted bounding box and the ground truth bounding box, reflecting the accuracy of the detected bounding box's localization. The edge IU / U further focuses on the target edge region, used to evaluate the degree of overlap between the detected bounding box edge and the ground truth edge, which is particularly important for tasks sensitive to edge details. Edge pixel accuracy is an indicator that measures the degree of matching between the detected bounding box edge pixels and the ground truth edge pixels, directly reflecting the fineness of edge detection. Target detection accuracy provides an overall assessment, edge IU / U optimizes edge localization, and edge pixel accuracy improves the reliability of edge details, collectively affecting the overall performance and practicality of target detection. The edge accuracy parameters are correlated with the edge accuracy value in the segmentation stage, specifically as follows: The higher the target detection accuracy value, the less closely the detected bounding box is to the ground truth target, resulting in lower initial localization accuracy in the subsequent segmentation stage, and thus lower edge accuracy values ​​in the segmentation stage. Edge intersection-union ratio (EUN) reflects the degree of overlap between the detected bounding box edge and the ground truth edge. A higher EUN value indicates more accurate edge localization, directly contributing to improved edge accuracy during segmentation. Edge pixel accuracy further refines this to the pixel level. High accuracy means that the detected edge pixels are highly consistent with the ground truth edge, providing finer edge information for segmentation and significantly improving segmentation edge accuracy.

[0034] It should be explained that the specific steps for determining whether to perform target detection accuracy optimization are as follows: if the target detection accuracy value is less than the lower bound of the detection accuracy reference, then target detection accuracy optimization is not performed; if the target detection accuracy value is within the detection accuracy reference range, then target detection accuracy optimization is performed; if the target detection accuracy value is greater than the upper bound of the detection accuracy reference, then an early warning message is triggered to notify relevant personnel.

[0035] It's important to note that dynamically adjusting the bounding box rotation angle based on false detection density per unit image area is a feasible approach to object detection optimization. Its core logic involves dynamically correcting the bounding box rotation angle by analyzing the distribution density of false detections in the image in real time, thereby reducing the false detection rate and improving detection accuracy. False detection density per unit image area refers to the number of bounding boxes incorrectly detected as targets within a unit area of ​​the image (e.g., per square pixel, per square centimeter). Specifically, it's the ratio of the total number of incorrectly detected bounding boxes to the total image area. This metric reflects the "density" of false detections in the image; higher density indicates a more severe false detection problem (potentially due to target overlap, complex backgrounds, angular deviations, etc.). The bounding box rotation angle (usually referring to its angle with the horizontal direction) is a key parameter describing the target's pose in object detection (e.g., tilted text, rotating vehicles, etc.). When the false detection density is high, it may be due to inaccurate bounding box angle estimation, leading to incorrect selection of background or similar targets. Therefore, by analyzing the spatial distribution of false detection density (e.g., a region with high false detection density), the angular deviation of the bounding boxes within that region can be inferred, and the angle can be dynamically corrected to reduce false detections.

[0036] The specific steps for dynamically adjusting the bounding box rotation angle are as follows: If the false detection density per unit image area is greater than the upper limit of the false detection density setting, the result of harmonic averaging of the target detection accuracy value and the false detection density offset is obtained by querying the bounding box rotation angle mapping table to obtain the angle adjustment factor. This can specifically optimize the bounding box rotation angle, effectively reduce the impact of excessively high false detection density on detection accuracy, and thus improve the accuracy and reliability of target detection. The false detection density offset represents the difference between the false detection density per unit image area and the upper limit of the false detection density setting.

[0037] The angle adjustment factor is compared with its set value. If the angle adjustment factor is greater than or equal to the set value, the angle adjustment offset is retrieved from the bounding box rotation angle mapping table to obtain the clockwise bounding box rotation angle correction. This clockwise bounding box rotation angle correction is then applied to the current bounding box rotation angle to obtain the next bounding box rotation angle. This precise adjustment of the bounding box rotation angle further optimizes the target detection positioning accuracy, reduces detection errors caused by angle deviations, and thus enhances the overall performance of target detection. The angle adjustment offset represents the difference between the angle adjustment factor and its set value. The "addition" operation represents an addition operation.

[0038] If the angle adjustment factor is less than the set value, the angle adjustment deviation is retrieved from the bounding box rotation angle mapping table to obtain the counter-clockwise bounding box rotation angle correction. Based on the current bounding box rotation angle, the counter-clockwise bounding box rotation angle correction is subtracted to obtain the next bounding box rotation angle. This allows for targeted and precise reverse adjustment of the bounding box rotation angle, effectively compensating for angle deviations, further improving the positioning accuracy of target detection, reducing detection errors caused by angle issues, and enhancing the overall performance of target detection. The angle adjustment deviation represents the difference between the angle adjustment factor and the set value. The subtraction operation is used to represent the difference.

[0039] In this embodiment, the false detection density offset is dynamically calculated by comparing the false detection density per unit image area with the upper bound of the false detection density. This offset is then harmonicly averaged with the target detection accuracy value. An angle adjustment factor is queried from the bounding box rotation angle mapping table to achieve intelligent control of the bounding box rotation angle. When the angle adjustment factor is greater than or equal to the set value, clockwise rotation angle correction is applied. When it is less than the set value, counterclockwise rotation angle correction is applied to adjust the difference. This accurately adjusts the bounding box rotation angle, effectively reducing detection errors caused by excessively high false detection density or angle deviation. This improves the positioning accuracy and robustness of target detection, enhances the system's adaptability to complex scenes, provides more accurate target area input for subsequent vehicle damage area segmentation, and ultimately improves the overall accuracy and reliability of vehicle damage recognition.

[0040] Dynamic control of bounding box rotation angle also includes: If the false detection density per unit image area is within the set false detection density range, dynamic adjustment of the bounding box rotation angle will not be performed. This avoids unnecessary angle adjustment operations when the false detection density is within a reasonable range, reduces the system's computational load, and maintains the stability of the current bounding box rotation angle. This complements the precise adjustment when the false detection density exceeds the limit, jointly ensuring the efficiency and accuracy of target detection under different false detection densities, and further optimizing the overall detection performance. The set false detection density range represents the closed interval formed by the lower and upper bounds of the set false detection density.

[0041] If the false detection density per unit image area is less than the lower bound of the false detection density setting, the result of harmonic averaging the target detection accuracy value and the false detection density deviation is retrieved from the bounding box rotation angle mapping table to obtain the angle adjustment factor. This provides a basis for targeted bounding box rotation angle adjustment, forming a complete response system with the adjustment when the false detection density is not adjusted within the set range and exceeds the upper bound. This ensures that detection can be optimized in a reasonable way under different false detection density conditions, further improving the adaptability and accuracy of target detection. The false detection density deviation represents the difference between the false detection density per unit image area and the lower bound of the false detection density setting.

[0042] Based on the comparison between the angle adjustment factor and the set value of the angle adjustment factor, if the angle adjustment factor is greater than or equal to the set value, the angle adjustment offset is retrieved from the bounding box rotation angle mapping table to obtain the counterclockwise rotation angle compensation amount of the bounding box. The current rotation angle of the bounding box is superimposed based on the counterclockwise rotation angle compensation amount to obtain the next bounding box rotation angle. This can accurately achieve counterclockwise angle compensation of the bounding box in scenarios with low false detection density. In conjunction with the adjustment methods under other false detection density conditions, the angle adjustment is further refined, improving the positioning accuracy and overall stability of target detection. The angle adjustment offset represents the difference between the angle adjustment factor and the set value of the angle adjustment factor.

[0043] If the angle adjustment factor is less than the set value, the angle adjustment deviation is retrieved from the bounding box rotation angle mapping table to obtain the clockwise bounding box rotation angle compensation. Based on the current bounding box rotation angle, the clockwise bounding box rotation angle compensation is processed to obtain the next bounding box rotation angle. This can accurately achieve clockwise angle compensation of the bounding box in scenarios with low false detection density. Combined with other control methods for different false detection density states and angle adjustment factors, the angle adjustment system is further improved, enhancing the positioning accuracy and overall stability of target detection. The angle adjustment deviation represents the difference between the angle adjustment factor and the set value.

[0044] In this embodiment, a complete dynamic angle control system is constructed by dividing the false detection density per unit image area into three states: below a set lower bound, within a set range, and above a set upper bound. When the false detection density is within the set range, the system does not perform angle adjustment to reduce computational load and maintain detection stability. When the false detection density is below the set lower bound, the angle control factor is queried based on the harmonic average of the deviation between the target detection accuracy value and the false detection density. By comparing it with the set value, counterclockwise or clockwise rotation compensation is performed respectively to achieve precise fine-tuning of the bounding box angle. When the false detection density is above the set upper bound, clockwise or counterclockwise angle correction is performed through a similar mechanism. This multi-level and differentiated control strategy not only effectively addresses different false detection density scenarios and improves the positioning accuracy and robustness of target detection, but also optimizes system operating efficiency by avoiding unnecessary adjustments, providing more efficient, accurate, and stable technical support for vehicle damage recognition.

[0045] It should also be noted that the reference upper and lower limits for the direction error recovery time are dynamically adjusted based on the direction error accumulation rate. The direction error accumulation rate describes the proportion of frames in a series of consecutive frames where the direction deviation angle exceeds a threshold, essentially reflecting the frequency and range of the deviation. The direction error recovery time describes the average time for a single deviation to recover from exceeding the limit, reflecting the duration of a single deviation. Both together characterize the "persistence characteristic" of the direction deviation: the accumulation rate focuses on the "overall persistence ratio across multiple frames," while the recovery time focuses on the "duration of a single event." This correlation provides a basis for adjusting the recovery time threshold using the accumulation rate—judging the overall severity of the deviation through the accumulation rate, and then dynamically adapting a reasonable threshold range for the recovery time.

[0046] like Figure 2 The diagram shown is a flowchart of the dynamic adjustment of the direction error recovery time reference interval for an intelligent vehicle damage detection and identification data processing method provided in this application embodiment. The specific logic is as follows: The dynamic adjustment of the direction error recovery time reference interval involves the following steps: If the cumulative direction error rate is greater than the upper limit of the cumulative rate reference, the target detection accuracy value and the cumulative rate offset are retrieved and matched from the direction error recovery time mapping table to obtain the upper limit of the direction error recovery time reference by adjustment, thus obtaining the next upper limit of the direction error recovery time reference. If the cumulative direction error rate is within the cumulative rate reference interval, the dynamic adjustment of the direction error recovery time reference interval is not performed. If the cumulative direction error rate is less than the lower limit of the cumulative rate reference, the target detection accuracy value and the cumulative rate deviation are retrieved and matched from the direction error recovery time mapping table to obtain the lower limit of the direction error recovery time reference by adjustment. Based on the current lower limit of the direction error recovery time reference, the lower limit of the direction error recovery time reference is adjusted by difference to obtain the next lower limit of the direction error recovery time reference.

[0047] If the cumulative rate of directional error exceeds the reference upper limit of the cumulative rate, the target detection accuracy value and the cumulative rate offset are retrieved and matched from the directional error recovery time mapping table to obtain the upward adjustment of the reference upper limit of the directional error recovery time. Based on the upward adjustment of the reference upper limit of the directional error recovery time, the current reference upper limit of the directional error recovery time is superimposed to obtain the next reference upper limit of the directional error recovery time. This can specifically extend the reference upper limit of the recovery time when the directional error accumulation exceeds the limit, providing a more reasonable time benchmark for the effective recovery of directional error. In conjunction with the control methods in other scenarios, it can further improve the control capability of directional error and ensure the stability and accuracy of target detection. The cumulative rate offset represents the difference between the cumulative rate of directional error and the reference upper limit of the cumulative rate.

[0048] If the cumulative rate of direction error is within the cumulative rate reference range, dynamic adjustment of the direction error recovery time reference range will not be performed. This avoids unnecessary adjustment operations and reduces system resource consumption when the cumulative rate of direction error is within a reasonable range. At the same time, it maintains the stability of the direction error recovery time reference range, complementing the upward adjustment when the cumulative rate of direction error exceeds the upper limit. Together, they ensure efficient recovery under different directional error accumulation states and further optimize overall performance. The cumulative rate reference range represents the closed interval formed by the lower limit and upper limit of the cumulative rate reference.

[0049] If the cumulative rate of directional error is less than the reference lower limit of the cumulative rate, the target detection accuracy value and the cumulative rate deviation are retrieved and matched from the directional error recovery time mapping table to obtain the adjustment amount of the reference lower limit of the directional error recovery time. Based on the current reference lower limit of the directional error recovery time, the adjustment amount of the reference lower limit of the directional error recovery time is processed to obtain the next reference lower limit of the directional error recovery time. This allows for targeted adjustment of the reference lower limit of the recovery time when the cumulative rate of directional error is low. This forms a complete system with the no adjustment when the cumulative rate is within the reference range and the upward adjustment when it exceeds the upper limit. This ensures the accuracy and efficiency of directional error recovery under different cumulative states and further optimizes the overall performance. The cumulative rate deviation represents the difference between the reference lower limit of the cumulative rate and the cumulative rate of directional error.

[0050] In this embodiment, a dynamic adjustment mechanism for the bounding box rotation angle based on hierarchical control of false detection density is established. When the false detection density is within a reasonable range, the angle remains stable to reduce computational overhead. When the false detection density is too high or too low, precise clockwise or counterclockwise angle compensation and correction are performed respectively to achieve dynamic optimization of the target detection angle deviation. This significantly improves the positioning accuracy, robustness, and adaptability of vehicle damage target detection, providing higher quality target area input for subsequent damage area segmentation and recognition. Overall, the performance and practicality of the vehicle damage recognition system are optimized.

[0051] The third step of this intelligent vehicle damage identification data processing method is to determine whether to perform edge accuracy optimization in the segmentation stage based on the edge accuracy value in the segmentation stage. If yes, the vehicle damage identification result is output after the edge accuracy optimization in the segmentation stage; otherwise, the vehicle damage identification result is output directly. Edge accuracy optimization in the segmentation stage includes dynamic control of boundary symmetry distance and dynamic control of the edge root mean square error qualification threshold.

[0052] Furthermore, the specific steps for determining whether to perform edge precision optimization in the segmentation stage are as follows: if the edge precision value in the segmentation stage is greater than or equal to the edge precision reference value, then edge precision optimization in the segmentation stage is not performed; if the edge precision value in the segmentation stage is less than the edge precision reference value, then edge precision optimization in the segmentation stage is performed. Edge precision optimization in the segmentation stage includes determining whether to perform dynamic adjustment of the boundary symmetry distance based on the boundary offset error. If so, then after performing dynamic adjustment of the boundary symmetry distance, determining whether to perform dynamic adjustment of the edge root mean square error qualification threshold.

[0053] It needs to be explained that the boundary symmetry distance is dynamically adjusted based on the boundary offset error. The boundary offset error is defined as "the average Euclidean distance between the segmented edge pixels and the corresponding real edge pixels," which essentially reflects the unidirectional average offset of the segmented edge relative to the real edge (emphasizing the average value of the overall offset). The boundary symmetry distance is calculated by averaging the maximum values ​​of the distance from the segmented edge to the real edge and the distance from the real edge to the segmented edge, comprehensively measuring the bidirectional offset magnitude of the edge (emphasizing the impact of extreme values ​​in the bidirectional offset on the overall result). Both revolve around the "spatial deviation between the segmented edge and the real edge." The former reflects the average level of the offset, while the latter reflects the bidirectional extreme characteristics of the offset. This complementarity provides the basis for "adjusting the boundary symmetry distance using the boundary offset error"—judging the severity of the overall offset through the boundary offset error and dynamically adapting the calculation logic or threshold standard of the boundary symmetry distance.

[0054] like Figure 3 The diagram shows a flowchart of the dynamic adjustment of boundary symmetry distance in an intelligent vehicle damage detection and identification data processing method provided in this application embodiment. The specific logic is as follows: If the boundary offset error is less than the lower limit of the offset error reference, the edge precision correction amount and the offset error deviation amount are retrieved and matched from the symmetry distance mapping table to obtain the boundary symmetry distance reduction amount. Based on the current boundary symmetry distance, the boundary symmetry distance reduction amount is adjusted to obtain the next boundary symmetry distance. If the boundary offset error is within the offset error reference range, the dynamic adjustment of boundary symmetry distance is not performed. If the boundary offset error is greater than the upper limit of the offset error reference, the edge precision correction amount and the offset error deviation amount are retrieved and matched from the symmetry distance mapping table to obtain the boundary symmetry distance increase amount. Based on the boundary symmetry distance increase amount, the current boundary symmetry distance is superimposed to obtain the next boundary symmetry distance.

[0055] As further explained, the decision to perform dynamic adjustment of the boundary symmetry distance is based on the boundary offset error. The specific adjustment steps are as follows: If the boundary offset error is less than the lower limit of the offset error reference, the edge precision correction amount and the offset error deviation amount are retrieved and matched from the symmetry distance mapping table to obtain the boundary symmetry distance reduction amount. Based on the current boundary symmetry distance, the boundary symmetry distance reduction amount is adjusted to obtain the next boundary symmetry distance. The edge precision correction amount represents the difference between the edge precision reference value and the edge precision value in the segmentation stage. At this time, the average Euclidean distance between the segmented edge and the real edge is extremely small, and the system is in a high-precision operating state. To further improve the fineness of the segmented edges, the qualified threshold of the boundary symmetry distance needs to be lowered. This adjustment makes the evaluation standard of the boundary symmetry distance more stringent and more sensitive to subtle bidirectional offsets, thereby timely identifying potential local bidirectional offset problems and avoiding the masking of local detail errors due to excessively good overall offset. Therefore, the boundary symmetry distance is reduced. The offset error deviation amount represents the difference between the lower limit of the offset error reference and the boundary offset error. Here, the reduction processing refers to the subtraction operation.

[0056] If the boundary offset error is within the offset error reference range, dynamic adjustment of the boundary symmetry distance will not be performed. In this case, the average offset degree is within the preset reasonable range, the system is operating normally, and the correlation between the overall offset and the bidirectional offset meets expectations, so as to reflect the extreme values ​​of the bidirectional offset in a balanced manner, without making additional relaxation or tightening of the bidirectional offset. This is done to ensure the stability of the evaluation, avoid misjudging normal offsets due to over-adjustment, and at the same time take into account the monitoring capability of abnormal bidirectional offsets. The offset error reference range represents the closed interval formed by the lower limit and upper limit of the offset error reference.

[0057] If the boundary offset error exceeds the upper limit of the offset error reference, the edge precision correction amount and the offset error offset amount are retrieved and matched from the symmetric distance mapping table to obtain the boundary symmetric distance adjustment amount. The current boundary symmetric distance is then superimposed based on this adjustment amount to obtain the next boundary symmetric distance. At this point, the average offset between the segmented edge and the real edge far exceeds the reasonable range, and the system is in a state of accuracy fluctuation. Adjusting the boundary symmetric distance significantly increases its acceptable threshold, directly expanding the numerical range of the boundary symmetric distance and allowing for a wider range of bidirectional offsets. Simultaneously, by reducing the weighting coefficient of the "maximum bidirectional distance" in the average calculation, the impact of extreme bidirectional offsets on the overall evaluation result is reduced. This further adapts the calculation logic to scenarios with severe overall offsets, avoiding misjudgments due to overly strict standards. This adjustment allows the boundary symmetric distance to better accommodate the associated deviations caused by the overall offset, leaving room for the system to correct the overall offset. The offset error represents the difference between the boundary offset error and the upper limit of the offset error reference.

[0058] In this embodiment, adaptive optimization of the offset between the segmented edge and the real edge is achieved by dynamically adjusting the boundary symmetry distance. When the boundary offset error is extremely small, the system reduces the acceptable threshold for the boundary symmetry distance to enhance sensitivity to local bidirectional offsets and prevent detailed errors from being masked. When the offset error is within a reasonable range, the system maintains a stable evaluation standard, balancing anomaly monitoring and operational stability. When the offset error is too large, the system increases the boundary symmetry distance and reduces the weight of extreme offsets to accommodate overall deviations and avoid misjudgments. This mechanism effectively improves the refinement quality and robustness of the segmented edges, ensuring high accuracy and reliability in vehicle damage identification.

[0059] As a further explanation, the specific control steps for determining whether to perform dynamic adjustment of the marginal root mean square error qualification threshold are as follows: If the average deviation distance of edge pixels is less than the lower limit of the average deviation distance reference, the edge precision correction amount and the average deviation distance offset are retrieved and matched from the average deviation distance mapping table to obtain the reduction amount of the edge root mean square error qualified threshold. Based on the edge root mean square error qualified threshold, the reduction amount of the edge root mean square error qualified threshold is adjusted downward to obtain the current edge root mean square error qualified threshold. This allows for targeted reduction of the error qualified threshold when the edge pixel deviation is small, strictly controlling the edge detection accuracy. This, combined with adjustments under other deviation levels, further improves the accuracy and reliability of edge detection. The average deviation distance offset represents the difference between the lower limit of the average deviation distance reference and the average deviation distance of the edge pixels. The downward adjustment refers to a subtraction operation.

[0060] If the average deviation distance of edge pixels is within the reference range of average deviation distance, dynamic adjustment of the acceptable threshold of the root mean square error of the edge will not be performed. This can avoid unnecessary threshold adjustments when the deviation of edge pixels is reasonable, reduce the computational burden of the system, and maintain the stability of the current acceptable threshold of error. This complements the strict control when the average deviation distance of edge pixels is less than the lower limit of the reference. Together, they ensure the accuracy and efficiency of edge detection under different deviation levels, and further optimize the edge detection performance. The reference range of average deviation distance represents the closed interval formed by the lower limit of the reference range of average deviation distance and the upper limit of the reference range of average deviation distance.

[0061] If the average deviation distance of edge pixels is greater than the upper limit of the average deviation distance reference, the edge precision correction amount and the average deviation distance deviation amount are retrieved and matched from the average deviation distance mapping table to obtain the increase amount of the edge root mean square error qualified threshold. Based on the edge root mean square error qualified threshold, the increase amount of the edge root mean square error qualified threshold is adjusted upward to obtain the current edge root mean square error qualified threshold. The error qualified threshold is increased in a targeted manner to flexibly adapt to the detection requirements. This works in conjunction with the stable control of the average deviation distance of edge pixels within the reference range and the strict control when it is less than the reference lower limit, to fully ensure the accuracy and adaptability of edge detection under different deviation degrees. The average deviation distance deviation amount represents the difference between the average deviation distance of edge pixels and the upper limit of the average deviation distance reference.

[0062] It's important to explain that the dynamic adjustment of the root mean square error (RMSE) threshold based on the average deviation distance of edge pixels is crucial. The average deviation distance (which can be understood as the average distance between an edge pixel and a reference, similar to the "average" characteristic of boundary offset error) reflects the overall deviation of edge pixels from the reference, emphasizing the "central tendency of the overall offset." The RMSE, calculated as the square root of the average of the squared distances of edge pixels from the fitted curve, essentially reflects the dispersion of edge pixels from the fitted curve and the impact of extreme deviations. Both revolve around the "degree of deviation of edge pixels," with the former reflecting the average offset and the latter reflecting the dispersion and impact of extreme values. This complementarity provides the basis for "adjusting the RMSE using the average deviation distance"—judging the severity of the overall offset by the average deviation distance and dynamically adapting the RMSE threshold.

[0063] In this embodiment, adaptive optimization of the average deviation distance of edge pixels is achieved by dynamically adjusting the root mean square error (RMSE) threshold. When the deviation distance is extremely small, the system lowers the error threshold to strictly control edge accuracy; when the deviation distance is within a reasonable range, the threshold is kept stable to reduce computational burden; when the deviation distance is too large, the error threshold is increased to flexibly adapt to detection requirements. The synergistic effect of these three factors effectively improves the accuracy, reliability, and adaptability of edge detection, ensuring refined edge recognition performance under different degrees of deviation.

[0064] like Figure 4The diagram shows a schematic of an intelligent vehicle damage detection and identification data processing system provided in this application embodiment. This system includes: a detection accuracy evaluation module, a target detection accuracy optimization and edge accuracy evaluation module, and a segmentation stage edge accuracy optimization module. The detection accuracy evaluation module acquires vehicle image data and performs target detection on the vehicle image to obtain detection accuracy parameters. Based on these parameters, it obtains a target detection accuracy value to evaluate the accuracy of identifying vehicle damage during the target detection process. The target detection accuracy optimization and edge accuracy evaluation module determines whether to perform target detection accuracy optimization based on the target detection accuracy value. If so, it performs vehicle damage region segmentation after target detection accuracy optimization; otherwise, it directly performs vehicle damage region segmentation and acquires edge accuracy parameters during the segmentation process. Based on these parameters, it obtains a segmentation stage edge accuracy value to evaluate the matching accuracy between the predicted vehicle damage edge and the actual vehicle damage edge. The segmentation stage edge accuracy optimization module determines whether to perform segmentation stage edge accuracy optimization based on the segmentation stage edge accuracy value. If so, it outputs the vehicle damage identification result after segmentation stage edge accuracy optimization; otherwise, it directly outputs the vehicle damage identification result.

[0065] In this embodiment, modular collaborative control enables adaptive optimization of accuracy throughout the entire vehicle damage recognition process. The detection accuracy evaluation module first acquires a vehicle image and performs target detection, generating a target detection accuracy value to assess the accuracy of damage recognition. The target detection accuracy optimization and edge accuracy evaluation module determines whether to optimize based on this accuracy value, then performs vehicle damage region segmentation and extracts edge accuracy parameters based on the segmentation results to obtain the edge accuracy value for the segmentation stage, used to evaluate the fit between the predicted and real edges. The segmentation stage edge accuracy optimization module then determines whether to further optimize the edges based on the edge accuracy value, ultimately outputting a high-precision vehicle damage recognition result. Through step-by-step judgment and dynamic adjustment, each module ensures timely optimization of the detection and segmentation process when accuracy is insufficient, and efficient output when requirements are met, significantly improving the accuracy and practicality of vehicle damage recognition.

[0066] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent vehicle damage detection and identification data processing method, characterized in that, Includes the following steps: Vehicle image data is acquired, and target detection is performed on the vehicle images to obtain detection accuracy parameters. Based on the detection accuracy parameters, a target detection accuracy value is obtained to evaluate the accuracy of identifying vehicle damage during the target detection process. The target detection accuracy value determines whether to perform target detection accuracy optimization. If yes, the vehicle damage area segmentation process is performed after target detection accuracy optimization. If no, the vehicle damage area segmentation process is performed directly, and the edge accuracy parameters in the vehicle damage area segmentation process are obtained. The edge accuracy value of the segmentation stage is obtained based on the edge accuracy parameters to evaluate the matching accuracy between the predicted vehicle damage edge and the real vehicle damage edge. The target detection accuracy optimization includes dynamic adjustment of the bounding box rotation angle and dynamic adjustment of the direction error recovery time reference interval. Whether to perform edge precision optimization in the segmentation stage is determined based on the edge precision value in the segmentation stage. If yes, the vehicle damage recognition result is output after the edge precision optimization in the segmentation stage. If no, the vehicle damage recognition result is output directly. The edge precision optimization in the segmentation stage includes dynamic adjustment of boundary symmetry distance and dynamic adjustment of the edge root mean square error qualification threshold.

2. The intelligent vehicle damage detection and identification data processing method according to claim 1, characterized in that, The specific steps for obtaining the target detection accuracy value include: The detection accuracy parameters include bounding box offset, damage range error rate, and minimum bounding box distance; The bounding box offset influence value is obtained by combining the results of the relative deviation ratio between the bounding box offset and the set value of the bounding box offset compensation factor. The error rate impact value is obtained by combining the results of the relative deviation ratio between the damage range error rate and the damage range error rate set value by the error rate compensation factor. The minimum distance influence value is obtained by combining the results of processing the minimum distance of the bounding box and the ratio of the minimum distance setting value of the bounding box by the minimum distance compensation factor. By coupling the bounding box offset influence value, the error rate influence value, and the minimum distance influence value, the target detection accuracy value is obtained. The specific steps for determining whether to perform target detection accuracy optimization are as follows: If the target detection accuracy value is less than the lower bound of the detection accuracy reference, then target detection accuracy optimization will not be performed; If the target detection accuracy value is within the detection accuracy reference range, then target detection accuracy optimization is performed; If the target detection accuracy value is greater than the upper limit of the detection accuracy reference, an early warning message will be triggered to notify relevant personnel.

3. The intelligent vehicle damage detection and identification data processing method according to claim 1, characterized in that, The dynamic adjustment of the bounding box rotation angle is achieved through the following steps: If the false detection density per unit image area is greater than the upper limit of the false detection density setting, the angle adjustment factor is obtained by querying the bounding box rotation angle mapping table by averaging the target detection accuracy value and the false detection density offset. The false detection density offset represents the degree of positive deviation between the false detection density per unit image area and the upper limit of the false detection density setting. The angle adjustment factor is compared with the set value. If the angle adjustment factor is greater than or equal to the set value, the angle adjustment offset is retrieved from the bounding box rotation angle mapping table to obtain the clockwise rotation angle correction amount of the bounding box. The current rotation angle of the bounding box is then superimposed on the clockwise rotation angle correction amount to obtain the next bounding box rotation angle. The angle adjustment offset represents the degree of positive deviation between the angle adjustment factor and the set value. If the angle adjustment factor is less than the set value of the angle adjustment factor, the angle adjustment deviation is retrieved from the bounding box rotation angle mapping table to obtain the counterclockwise rotation angle correction amount of the bounding box. Based on the current rotation angle of the bounding box, the counterclockwise rotation angle correction amount of the bounding box is processed by difference to obtain the next bounding box rotation angle. The angle adjustment deviation amount represents the degree of negative deviation between the angle adjustment factor and the set value of the angle adjustment factor.

4. The intelligent vehicle damage detection and identification data processing method according to claim 1, characterized in that, The dynamic control of the bounding box rotation angle also includes: If the false detection density per unit image area is within the set range of false detection density, then dynamic adjustment of the bounding box rotation angle will not be performed. The set range of false detection density represents the closed interval formed by the lower bound of the set false detection density and the upper bound of the set false detection density. If the false detection density per unit image area is less than the lower bound of the false detection density setting, the angle adjustment factor is obtained by querying the bounding box rotation angle mapping table by the result of harmonic averaging the target detection accuracy value and the false detection density deviation. The false detection density deviation represents the degree of negative deviation between the false detection density per unit image area and the lower bound of the false detection density setting. Based on the comparison between the angle adjustment factor and the angle adjustment factor setting value, if the angle adjustment factor is greater than or equal to the angle adjustment factor setting value, the angle adjustment offset is retrieved from the bounding box rotation angle mapping table to obtain the counterclockwise rotation angle compensation amount of the bounding box. The current rotation angle of the bounding box is superimposed based on the counterclockwise rotation angle compensation amount to obtain the next bounding box rotation angle. The angle adjustment offset represents the degree of positive deviation between the angle adjustment factor and the angle adjustment factor setting value. If the angle adjustment factor is less than the angle adjustment factor setting value, the angle adjustment deviation is retrieved from the bounding box rotation angle mapping table to obtain the clockwise rotation angle compensation amount of the bounding box. Based on the current rotation angle of the bounding box, the clockwise rotation angle compensation amount of the bounding box is processed by difference to obtain the next bounding box rotation angle. The angle adjustment deviation amount represents the degree of negative deviation between the angle adjustment factor and the angle adjustment factor setting value.

5. The intelligent vehicle damage detection and identification data processing method according to claim 1, characterized in that, The dynamic adjustment of the direction error recovery time reference interval is carried out through the following steps: If the cumulative rate of directional error is greater than the upper limit of the cumulative rate reference, the target detection accuracy value and the cumulative rate offset are retrieved and matched from the directional error recovery time mapping table to obtain the upper limit of the directional error recovery time reference. The current upper limit of the directional error recovery time reference is superimposed based on the upper limit of the directional error recovery time reference to obtain the next upper limit of the directional error recovery time reference. The cumulative rate offset represents the degree of deviation between the cumulative rate of directional error and the upper limit of the cumulative rate reference. If the cumulative rate of direction error is within the cumulative rate reference interval, then the dynamic adjustment of the reference interval of direction error recovery time will not be performed. The cumulative rate reference interval represents the closed interval formed by the lower limit of the cumulative rate reference and the upper limit of the cumulative rate reference. If the cumulative rate of directional error is less than the reference lower limit of the cumulative rate, the target detection accuracy value and the cumulative rate deviation are retrieved and matched from the directional error recovery time mapping table to obtain the adjustment amount of the reference lower limit of directional error recovery time. Based on the current reference lower limit of directional error recovery time, the adjustment amount of the reference lower limit of directional error recovery time is processed to obtain the next reference lower limit of directional error recovery time. The cumulative rate deviation represents the degree of deviation between the reference lower limit of directional error and the cumulative rate of directional error.

6. The intelligent vehicle damage detection and identification data processing method according to claim 1, characterized in that, The specific steps for obtaining the edge precision value at the segmentation stage are as follows: The edge accuracy parameters include target detection accuracy value, edge intersection-over-union ratio, and edge pixel accuracy. The detection accuracy impact value is obtained by combining the results of processing the target detection accuracy set value and the target detection accuracy value ratio by the detection accuracy compensation factor. The cross-union ratio influence value is obtained by combining the results of the edge cross-union ratio and the edge cross-union ratio set value processing by the cross-union ratio compensation factor. The accuracy impact value is obtained by combining the results of the accuracy compensation factor processing of the edge pixel accuracy and the ratio of the edge pixel accuracy set value. By coupling the impact values ​​of detection accuracy, cross-union ratio, and accuracy, the edge accuracy value of the segmentation stage is obtained.

7. The intelligent vehicle damage detection and identification data processing method according to claim 1, characterized in that, The specific steps for determining whether to perform edge precision optimization during the segmentation stage are as follows: If the edge precision value in the segmentation stage is greater than or equal to the edge precision reference value, then edge precision optimization in the segmentation stage will not be performed. If the edge precision value in the segmentation stage is less than the edge precision reference value, then edge precision optimization in the segmentation stage is performed. The edge precision optimization in the segmentation stage includes determining whether to perform dynamic adjustment of the boundary symmetry distance based on the boundary offset error. If so, then after performing dynamic adjustment of the boundary symmetry distance, it is determined whether to perform dynamic adjustment of the edge root mean square error qualification threshold.

8. The intelligent vehicle damage detection and identification data processing method according to claim 7, characterized in that, The specific steps for determining whether to perform dynamic adjustment of the boundary symmetry distance based on the boundary offset error are as follows: If the boundary offset error is less than the lower limit of the offset error reference, the edge precision correction amount and the offset error deviation amount are retrieved and matched from the symmetric distance mapping table to obtain the boundary symmetric distance reduction amount. Based on the current boundary symmetric distance, the boundary symmetric distance reduction amount is adjusted to obtain the next boundary symmetric distance. The edge precision correction amount represents the degree of deviation between the edge precision reference value and the edge precision value in the segmentation stage, and the offset error deviation amount represents the degree of deviation between the lower limit of the offset error reference and the boundary offset error. If the boundary offset error is within the offset error reference range, then dynamic adjustment of the boundary symmetry distance will not be performed. The offset error reference range refers to the closed interval formed by the lower limit of the offset error reference and the upper limit of the offset error reference. If the boundary offset error is greater than the upper limit of the offset error reference, the edge precision correction amount and the offset error offset amount are retrieved and matched from the symmetric distance mapping table to obtain the boundary symmetric distance adjustment amount. The current boundary symmetric distance is superimposed based on the boundary symmetric distance adjustment amount to obtain the next boundary symmetric distance. The offset error offset amount represents the degree of deviation between the boundary offset error and the upper limit of the offset error reference.

9. The intelligent vehicle damage detection and identification data processing method according to claim 7, characterized in that, The specific control steps for determining whether to perform dynamic adjustment of the marginal root mean square error qualification threshold are as follows: If the average deviation distance of edge pixels is less than the lower limit of the average deviation distance reference, the edge precision correction amount and the average deviation distance offset are retrieved and matched from the average deviation distance mapping table to obtain the reduction amount of the edge root mean square error qualified threshold. Based on the edge root mean square error qualified threshold, the reduction amount of the edge root mean square error qualified threshold is adjusted downward to obtain the current edge root mean square error qualified threshold. The average deviation distance offset represents the degree of deviation between the lower limit of the average deviation distance reference and the average deviation distance of edge pixels. If the average deviation distance of edge pixels is within the average deviation distance reference interval, then the dynamic adjustment of the edge root mean square error qualification threshold will not be performed. The average deviation distance reference interval represents the closed interval formed by the lower limit of the average deviation distance reference and the upper limit of the average deviation distance reference. If the average deviation distance of edge pixels is greater than the upper limit of the average deviation distance reference, the edge precision correction amount and the average deviation distance deviation amount are retrieved and matched from the average deviation distance mapping table to obtain the increase amount of the edge root mean square error qualified threshold. Based on the edge root mean square error qualified threshold, the increase amount of the edge root mean square error qualified threshold is adjusted upward to obtain the current edge root mean square error qualified threshold. The average deviation distance deviation amount represents the degree of deviation between the average deviation distance of edge pixels and the upper limit of the average deviation distance reference.

10. An intelligent vehicle damage detection and identification data processing system, wherein the intelligent vehicle damage detection and identification data processing system applies the intelligent vehicle damage detection and identification data processing method as described in any one of claims 1-9, characterized in that, It includes a detection accuracy evaluation module, a target detection accuracy optimization and edge accuracy evaluation module, and a segmentation stage edge accuracy optimization module: The detection accuracy evaluation module is used to acquire vehicle image data, perform target detection on the vehicle image to obtain detection accuracy parameters, and obtain target detection accuracy values ​​based on the detection accuracy parameters to evaluate the accuracy of identifying vehicle damage during the target detection process. The target detection accuracy optimization and edge accuracy evaluation module is used to determine whether to perform target detection accuracy optimization based on the target detection accuracy value. If yes, the vehicle damage area segmentation process is performed after the target detection accuracy optimization. If no, the vehicle damage area segmentation process is performed directly, and the edge accuracy parameters in the vehicle damage area segmentation process are obtained. The edge accuracy value of the segmentation stage is obtained based on the edge accuracy parameters to evaluate the matching accuracy between the predicted vehicle damage edge and the real vehicle damage edge. The segmentation stage edge precision optimization module is used to determine whether to perform segmentation stage edge precision optimization based on the segmentation stage edge precision value. If yes, the vehicle damage recognition result is output after the segmentation stage edge precision optimization; otherwise, the vehicle damage recognition result is output directly.

Images