Trailer service-based accident vehicle intelligent fixing method and device

Abstract

The invention discloses an accident vehicle intelligent fixing method and device based on trailer business, belongs to the technical field of trailer intelligent fixing, and aims to solve the problems that in existing trailer business, accident vehicle fixing processing is often judged through human experience, scientific loading and fixing of vehicles damaged due to special requirements are lacked, and the accident vehicle fixing processing efficiency is low. And secondary potential safety hazards exist in the transportation process of the accident vehicle. The method comprises the following steps: carrying out loading grade judgment under a related damage structure on a real-time three-dimensional architecture model of an accident vehicle and a standard vehicle model three-dimensional model to obtain a loading risk grade of the accident vehicle; the method comprises the following steps: performing damage characteristic identification under related multiple scales on a whole vehicle damage image set shot from multiple angles of an accident vehicle, and determining key damage structure data; performing damage feature-based factor influence contribution association processing on the key damage structure data to obtain structure safety condition data; and performing local area correction processing on the initial vehicle fixed position data to determine reference vehicle fixed position data.

Description

Technical Field

[0001] The present application relates to the field of intelligent trailer fixation, and in particular to an intelligent fixation method and device for accident vehicles based on trailer services. Background Art

[0002] As vehicles become more and more popular, the number of vehicles on the road is also increasing, and the probability of traffic accidents will also increase significantly. In some serious traffic accidents, the vehicle is severely damaged, making it unable to drive, or even severely deformed and unable to move, or the frame and body structure are severely damaged. In such cases, in order to avoid affecting normal traffic, towing services will be carried out. Tow trucks or flatbed trucks are used to tow the immobilized accident vehicle away as soon as possible so that subsequent repairs can be carried out.

[0003] In current towing operations, severely damaged or deformed accident vehicles are often manually secured to a flatbed truck using tools like chains and ropes. This leads to low securing efficiency (requiring collaboration among multiple personnel), poor stability (prone to loosening due to turbulence), and safety hazards (operators need close proximity to the damaged vehicle). In some cases involving severely damaged or significantly deformed vehicles, statistically analyzed cases show that manual securing can cause secondary damage (e.g., uneven force at the lashing points, causing deformation). Furthermore, manual securing is often based on judgment based on personal experience, often resulting in the securing of severely deformed vehicles.

[0004] Although there are adjustable pallets (such as expanding the main / auxiliary pallet size through a hydraulic controller) or wheel hub fixing devices (such as trailer locks with elastic connecting plates), none of them specifically address the special needs of damaged vehicles. For example, they cannot adapt to vehicles with deformed wheels or damaged suspensions; there is a lack of automatic compensation mechanism for vehicle center of gravity offset; the fixing process still requires manual assisted positioning, etc.

[0005] That is, the traditional method of fixing trailers for vehicles with severe deformation in accidents is generally based on human experience and judgment, and is combined with fixed fixing devices to complete the loading of the trailer. However, the current method of loading accident vehicles is often based on human experience and judgment, which is difficult for novice towing workers to get started. It is difficult to have better auxiliary positioning tools to achieve rapid fixation of vehicles with severe accidents, and unreasonable loading and fixation can easily cause secondary damage to vehicles during transportation. It basically relies on the human experience of the loader to determine the appropriate fixing point for the accident vehicle, and cannot provide "customized" safe fixation for severely deformed vehicles. There are certain safety hazards in the transportation of accident vehicles, which is not conducive to the safe and reasonable implementation of towing services. There is an urgent need for a method that can assist human loading and fixation to assist towing workers in completing the efficient and rapid fixation of vehicles with severe deformation and damage. Summary of the Invention

[0006] The embodiments of the present application provide an intelligent fixation method and device for accident vehicles based on towing services, which are used to solve the following technical problems: the fixation of accident vehicles in existing towing services is often based on human experience to determine the fixing position, which makes it difficult to reasonably and safely complete the loading and fixing of accident deformed vehicles. There is a lack of scientific loading and fixing for vehicles with special damage needs, and there will also be secondary safety hazards in the transportation of accident vehicles.

[0007] The embodiments of this application adopt the following technical solutions:

[0008] On the one hand, an embodiment of the present application provides an intelligent fixation method for an accident vehicle based on a towing service, comprising: identifying damage characteristics at multiple scales for a set of whole-vehicle damage images taken from multiple angles of the accident vehicle, and determining key damage structure data in each damage area; performing correlation processing on the key damage structure data in each damage area based on the contribution of factors affecting the damage characteristics, and evaluating and obtaining structural safety data for each damage area; performing a fixed force balance calculation on the accident vehicle based on a real-time three-dimensional architecture model to obtain initial vehicle fixed position data; performing local area correction processing on the initial vehicle fixed position data based on the entire vehicle body through the structural safety data of each damage area, determining reference vehicle fixed position data, and displaying the reference vehicle fixed position data in the form of a three-dimensional graph, thereby providing a reference for the fixation method of the accident vehicle.

[0009] The embodiments of the present application utilize multi-angle imaging and damage characteristic identification to more accurately assess vehicle damage, thereby improving the assessment of vehicle structural safety during the securing process. Furthermore, the loading risk level of the accident vehicle can be determined. In particular, for vehicles with high loading risk levels, more stringent securing measures can be implemented to reduce the risk of accidents. Deformable convolutional networks can also be used to identify damage characteristics at multiple scales, more accurately determining the key damaged structures within each damaged area, providing a basis for subsequent securing measures. The SHAP algorithm can also be used to evaluate the contribution of factors influencing the key damage structure data within each damaged area, enabling a more comprehensive assessment of the vehicle's structural safety data. Based on a real-time 3D structural model and structural safety data, the optimal vehicle securing position can be calculated to ensure optimal and stable securing results. Periodic monitoring of the trailer securing structure corresponding to the reference vehicle securing position data can also be performed to continuously generate loading and driving safety data, facilitating monitoring and maintenance of vehicle safety status. Damage identification, risk level assessment, and optimal securing position determination can also be automated, reducing manual intervention and improving work efficiency.

[0010] In a feasible embodiment, the method also includes the step of obtaining the real-time three-dimensional architecture model, which specifically includes: using a handheld visual sensor device of the staff to collect and process continuous image frames of the accident vehicle at all angles to obtain an accident vehicle image set and a corresponding visual odometry; wherein the visual odometry is the cumulative pose estimation position and motion trajectory during the image frame acquisition process; based on the visual odometry and the vehicle appearance features in the accident vehicle image set, an initial three-dimensional structure model of the accident vehicle is constructed; a stereo matching cost aggregation calculation is performed on the whole vehicle damage image set of the accident vehicle under the left and right visual differences to obtain a fitting matching cost contour curve of the whole vehicle damage image set; a global loop detection is performed on the fitting matching cost contour curve under the inter-frame pixel positioning change, and based on the constraint node parameters, a three-dimensional map of damage transformation points corresponding to the whole vehicle damage image set is constructed; based on the coordinate position of the same damage area in the whole vehicle damage image set, the three-dimensional map of damage transformation points is structurally fused with the initial three-dimensional structure model to obtain the real-time three-dimensional architecture model.

[0011] The embodiment of the present application improves the accuracy of collecting the damaged structural area by comparing the real-time three-dimensional model of the collected accident vehicle with the three-dimensional model of the standard vehicle model. Then, based on the corresponding non-overlapping area under the loss structure area, the scale comparison between the three-dimensional models is completed, and based on the volume of the non-overlapping area, the size of the actual loss area is determined. Only then can it be finally calculated whether the accident vehicle belongs to the high-risk loading level and whether to perform subsequent optimal fixed position point calculation.

[0012] In a feasible implementation, the damage characteristics at multiple scales are identified for the whole vehicle damage image set taken from multiple angles of the accident vehicle, and the key damage structure data in each damage area is determined, which specifically includes: transmitting the collected whole vehicle damage image set of the accident vehicle to a deformable convolutional network; wherein, the whole vehicle damage image set is a collection of images at several angles corresponding to each damage area; through the deformable convolutional network, the images in the damage image set under each damage area are uniformly resized to obtain damage images of the same size; the pre-acquired vehicle accident damage type image is processed to capture the key points of the damage type to determine the trend of graphic line changes; and the fuzzy features in the damage images of the same size are effectively processed based on the trend of graphic line changes. Robustness enhancement calculation under relevant graphic changes is performed to obtain a data-enhanced damage image; type recognition calculation under the deformation characteristics of vehicle structural damage is performed on the data-enhanced damage image to determine the damage characteristic data of each damage area; wherein, the damage characteristic data includes: intuitive damage characteristic data and hidden damage characteristic data; through a preset hybrid loss function, the damage characteristic data of each damage area is evaluated for the focus type under the key damage structure, and based on the positioning, segmentation and classification of the damage target, the key damage structure data in each damage area is identified and determined; wherein, the key damage structure data includes: dent damage, crack damage, wrinkle damage, perforation damage and structural loss damage; the hybrid loss function includes: classification loss, bounding box loss and mask loss.

[0013] The embodiment of the present application identifies damage characteristics at multiple scales on a set of vehicle damage images. That is, a deformable convolutional network is used to further analyze the collected damage structure images. Based on the trend of image line changes and the comprehensive recognition influence of intuitive damage characteristic data and hidden damage characteristic data, the damage targets under key damage structures are located, segmented, and classified. Finally, the key damage structure data in each damage area, that is, the detailed damage structure type, is identified and determined, which is conducive to clarifying the severity of the damage area and the specific damage type, and providing an accurate data basis for subsequent data evaluation and calculation.

[0014] In a feasible implementation, the data-enhanced damage image is subjected to type recognition calculations related to the deformation characteristics of vehicle structural damage to determine the damage characteristics of each damaged area, specifically comprising: performing feature capture processing related to dynamic convolution kernels and dynamic receptive fields on the data-enhanced damage image in the whole vehicle damage image, and performing secondary frame selection feature capture on the data-enhanced damage image based on an irregular vehicle damage shape template to obtain the intuitive damage characteristic data; performing high and low score marking processing on the fractional boundary overlapping boxes in the data-enhanced damage image, and performing hierarchical feature capture on all boundary boxes based on the marking score of each fractional boundary overlapping box to obtain the hidden damage characteristic data.

[0015] In a feasible embodiment, before performing factor contribution correlation processing on the key damage structure data in each damage area based on the damage characteristics, and evaluating and obtaining the structural safety status data of each damage area, the method also includes: performing feature capture processing on the data-enhanced damage image in the whole vehicle damage image with respect to the dynamic convolution kernel and the dynamic receptive field, and performing secondary frame selection feature capture on the data-enhanced damage image based on the irregular vehicle damage shape template to obtain the intuitive damage characteristic data; performing high and low score marking processing on the fractional boundary overlapping boxes in the data-enhanced damage image in the whole vehicle damage image, and performing hierarchical feature capture on all boundary boxes based on the marking score of each fractional boundary overlapping box to obtain the hidden damage characteristic data.

[0016] The embodiment of the present application ensures accurate classification of damages by marking the damage types of key damage structure data in each damage area one by one, which helps in subsequent detailed analysis. The numerical information corresponding to each damage type can also be recorded to provide basic data for quantitative analysis. The factor impact contribution value of each damage type is then calculated, which helps to understand the degree of influence of different damage types on the overall structural safety. At the same time, by performing a weighted average calculation of the marginal contribution under the feature combination through the SHAP algorithm, the impact of each factor on structural safety can be more accurately measured, even if there is interaction between these factors. Equally dividing the damage types based on the SHAP interaction value can more fairly distribute the impact contribution and avoid the situation where a single damage type may excessively affect the evaluation results. Finally, by obtaining the SHAP value of each damage area, the impact of each damage type on structural safety can be more comprehensively evaluated, providing a basis for the fixed strategy.

[0017] In a feasible implementation, the key damaged structural data in each damaged area are subjected to correlation processing based on the contribution of factors affecting the damage characteristics, and the structural safety status data of each damaged area are evaluated and obtained, specifically including: establishing a structural safety assessment prediction model; wherein, the structural safety assessment prediction model is used to perform comprehensive factor assessment processing on the key damaged structural data; through the structural safety assessment prediction model, the total SHAP value in each damaged area is subjected to the total contribution calculation of the relevant model score prediction value to obtain the percentage of accident damage factors in each damaged area; wherein, the percentage of accident damage factors is positively correlated with the total SHAP value; the percentage of accident damage factors is converted into a fractional value to obtain the structural safety status data of each damaged area; wherein, the higher the score in the structural safety status data, the more serious the corresponding damage degree.

[0018] The embodiment of the present application can conduct a comprehensive and integrated evaluation of the key damage structure data of the accident vehicle, taking into account a variety of influencing factors. The SHAP values ​​of each damage type in each damage area are added together to obtain a total SHAP value, which helps to identify the damage type that has the greatest impact on structural safety. Moreover, through interactive correlation analysis, the interaction relationship between different damage types is understood, which is crucial for understanding complex damage patterns and their impact on structural integrity. At the same time, the TAN Bayesian network structure is used to represent the factor effect relationship. This method can effectively capture the complex dependencies between variables and improve the predictive ability of the model. And based on the Bayesian network structure, the total contribution of the total SHAP value of each damage area is calculated, which helps to determine the importance of accident damage factors in structural safety. It helps to understand the degree of influence of various damage factors on the overall structural safety. Finally, through fractional numerical conversion, the percentage of accident damage factors is converted into structural safety data, making the evaluation results more intuitive and easy to understand.

[0019] In a feasible implementation, based on the real-time three-dimensional structural model, a fixed force balance calculation is performed on the accident vehicle to obtain initial vehicle fixed position data, specifically including: performing a body posture angle balance calculation on the real-time three-dimensional structural model to obtain body posture angle data; performing a static balance calculation on the real-time three-dimensional structural model regarding the single-point bearing capacity to obtain body single-point bearing capacity data; performing an automatic compensation calculation on the real-time three-dimensional structural model regarding the body center of gravity offset to obtain body center of gravity balance data; respectively determining the body posture angle data, the body single-point bearing capacity data, and the body center of gravity balance data as total constraint conditions; and performing a nonlinear force balance calculation on the real-time three-dimensional structural model regarding the body fixed points through the objective function corresponding to the total constraint conditions to obtain initial vehicle fixed position data of the real-time three-dimensional structural model.

[0020] In a feasible implementation, the structural safety data of each damaged area is used to perform a local area correction process on the initial vehicle fixed position data based on the entire vehicle body to determine the reference vehicle fixed position data, specifically including: performing a dynamic grading strategy process on the structural safety data to obtain a safety level area; wherein, the safety level area includes: a red level area, a yellow level area, and a green level area; performing a vehicle-wide discretization process on the real-time three-dimensional architecture model to establish an inter-node stiffness transfer function; according to the node positions corresponding to the safety level areas, the corresponding nodes in the inter-node stiffness transfer function are labeled to obtain labeled grid nodes; wherein, The annotated grid nodes contain three fixed meanings, and correspond to the colors of the security level areas; the initial vehicle fixed position data is loaded; a neighborhood scan of global grid nodes is performed on each initial point in the initial vehicle fixed position data, and based on the node avoidance attributes of the annotated grid nodes in the local area, the corresponding neighborhood nodes are subjected to node Euclidean distance calculation under the spherical correction domain to obtain a position offset vector for each initial point; the node avoidance attributes include: all avoidance attributes of red nodes and displacement compensation avoidance attributes of yellow nodes; all initial fixed points in the initial vehicle fixed position data are offset using the position offset vector to obtain the corrected reference vehicle fixed position data.

[0021] The embodiments of the present application use structural safety data to perform local area corrections on the vehicle's fixed position, ensuring a direct correlation between the fixed position and the vehicle's structural safety, thereby improving the safety of the fixed effect. The structural safety data is dynamically graded and divided into red, yellow, and green level areas, facilitating the rapid identification of areas with different safety risk levels. Furthermore, the entire vehicle is discretized using a NURBS surface mesh, which provides an accurate geometric representation and facilitates detailed mechanical analysis. Simultaneously, a stiffness transfer function is established between nodes, which helps simulate the mechanical properties of the vehicle structure and provides a mechanical basis for the correction of the fixed position. By labeling the grid nodes, clear identification is provided for areas with different safety levels, helping operators understand and implement the fixed strategy. The introduction of avoidance properties for red and yellow nodes, particularly where red nodes require complete avoidance and yellow nodes can be avoided with displacement compensation, enhances the flexibility of the fixed strategy. The initial vehicle fixed position data can also be scanned for the neighborhood of the global grid nodes and local corrections can be made based on the avoidance properties of the labeled grid nodes, which helps optimize the fixed position. Finally, by calculating the position offset vector of each initial point, the fixing position can be accurately adjusted to ensure that the vehicle structure will not suffer additional damage during the fixing process.

[0022] In a feasible embodiment, after the structural safety data of each damaged area is used to correct the initial vehicle fixed position data in a local area based on the entire vehicle body to determine the reference vehicle fixed position data, the method further includes: loading and fixing the accident vehicle onto a trailer based on the reference vehicle fixed position data; wherein the trailer fixing structure includes at least: a binding fixing structure, an auxiliary fixing buckle and a vehicle limiter; placing a pressure sensor at each fixed point in the reference vehicle fixed position data; periodically monitoring the pressure data of the trailer fixed structure through the pressure sensor, and feeding back and obtaining the trailer fixed structure feedback data; performing tabular data statistical processing on the trailer fixed structure feedback data to generate the loading and driving safety data for monitoring the accident vehicle during transportation.

[0023] On the other hand, an embodiment of the present application also provides an intelligent fixation device for an accident vehicle based on a towing service, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute an intelligent fixation method for an accident vehicle based on a towing service as described in any of the above embodiments.

[0024] This application provides a method and device for intelligently securing an accident vehicle based on a towing service. Compared with the prior art, the embodiments of this application have the following beneficial technical effects:

[0025] 1. Improved safety: Through multi-angle shooting and damage characteristic identification, the vehicle damage situation can be judged more accurately, thereby improving the assessment of vehicle structural safety during the fixing process.

[0026] 2. Risk level classification: It can judge the loading risk level of accident vehicles, especially for vehicles with high loading risk levels, so that more stringent fixing measures can be taken to reduce the risk of accidents.

[0027] 3. Damage characteristic identification: Using a deformable convolutional network to identify damage characteristics at multiple scales can more accurately determine the key damage structure in each damaged area, providing a basis for subsequent fixation measures.

[0028] 4. Structural safety assessment: The SHAP algorithm is used to evaluate the contribution of key damage structural data to each damage area, enabling a more comprehensive assessment of the vehicle's structural safety data.

[0029] 5. Intelligent fixing position determination: Based on real-time 3D architecture models and structural safety data, the optimal vehicle fixing position can be calculated to ensure the optimality and stability of the fixing effect.

[0030] 6. Periodic monitoring: Periodic monitoring of the trailer's fixed structure corresponding to the reference vehicle's fixed position data can continuously generate loading and driving safety data, facilitating the monitoring and maintenance of the vehicle's safety status.

[0031] 7. Improve work efficiency: It can automatically perform damage identification, risk level assessment and optimal fixing position determination, reducing manual intervention and improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0033] Figure 1 A flow chart of an intelligent method for fixing an accident vehicle based on a towing service provided in an embodiment of the present application;

[0034] Figure 2 A schematic structural diagram of a three-dimensional model of a standard vehicle model provided in an embodiment of the present application;

[0035] Figure 3 A schematic top view of a trailer flatbed provided in an embodiment of the present application in an optimal fixed position;

[0036] Figure 4 A schematic structural diagram of an intelligent fixing device for an accident vehicle based on a towing service provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0038] The embodiment of the present application provides an intelligent fixing method for accident vehicles based on towing services, such as Figure 1 As shown, the intelligent fixation method for an accident vehicle based on the towing service specifically includes steps S101-S106:

[0039] S101. Determine the loading level of the accident vehicle under the damaged structure using the real-time 3D structural model of the accident vehicle and the 3D model of the standard vehicle to obtain the loading risk level of the accident vehicle.

[0040] It should be noted that visual SLAM has a more advanced map form in the visual mapping process, that is, by constructing a three-dimensional animated map and being able to display models of all objects in the field of view, the robot is assisted in identification. Since the above-mentioned visual SLAM method is only a basic architecture, therefore, in order to further refine visual SLAM, this application designs a binocular visual SLAM system and a three-dimensional animation modeling method based on the SMA algorithm. First, the visual sensor uses the ZED2 binocular stereo camera of STEREOLABS, which can give full play to the advantages of the binocular camera and has better applicability in a wide range of application scenarios. Because the camera introduces various optical devices, distortion occurs during imaging, therefore, calibration is required, and the efficient and practical Zhang Zhengyou calibration method is adopted, which overcomes the shortcomings of the high-precision calibration objects required by the traditional calibration method, and only requires a printed chessboard to complete the calibration. Compared with self-calibration, it improves accuracy and is easy to operate.

[0041] Specifically, a modern visual SLAM algorithm and a handheld visual sensor device, which can be a dedicated industrial camera or a mobile phone camera app, are first used to capture and process continuous image frames of the accident vehicle from all angles, generating a set of accident vehicle images and the corresponding visual odometry. The visual odometry is the cumulative pose estimation and motion trajectory during the image frame acquisition process.

[0042] Furthermore, based on the visual odometry and the vehicle appearance features in the accident vehicle image set, an initial three-dimensional structural model of the accident vehicle is constructed.

[0043] In one embodiment, towing operators carry handheld visual sensor devices, such as cameras or other visual sensors. They then use the visual sensors to capture continuous image frames from all angles of the accident vehicle. These captured image frames are then preprocessed, such as through noise reduction and color correction. Modern visual SLAM algorithms are then applied to extract feature points from the image frames and calculate the pose changes between consecutive frames to obtain a visual odometry. To construct a 3D structural model, the vehicle's external features must first be extracted from the image set. Based on the visual odometry and vehicle external features, an initial 3D structural model of the accident vehicle can then be constructed.

[0044] Furthermore, a preset binocular camera is used to perform stereo matching cost aggregation calculation on the whole vehicle damage image set under left and right visual difference to obtain a fitting matching cost contour curve of the whole vehicle damage image set.

[0045] Furthermore, a global loop detection is performed on the fitted matching cost contour curve under the change of pixel positioning between frames, and a three-dimensional damage transformation point map corresponding to the whole vehicle damage image set is constructed based on the constraint node parameters.

[0046] In one embodiment, a pre-set binocular camera system is also used. Stereo matching is then performed on the full-vehicle damage image set of the accident vehicle under left-right visual disparity. The stereo matching cost is then calculated and aggregated to obtain a fitted matching cost contour curve. Global loop detection is then performed on the fitted matching cost contour curve to identify inter-frame pixel positioning changes. Finally, based on the constraint node parameters, a three-dimensional map of the damage transformation points corresponding to the full-vehicle damage image set is constructed.

[0047] As a feasible implementation, after the above process is completed, stereo matching is required. This is an extremely intensive task, requiring the disparity of nearly all pixels in the left view image to be obtained based on the corresponding points of the left and right binoculars, thereby obtaining a dense disparity map. The SMA algorithm modeling process used in this study mainly consists of four parts: matching cost calculation, matching cost aggregation, disparity calculation and optimization, and disparity refinement. First, the stereo matching cost calculation part essentially calculates the grayscale similarity of the disparity between the two images. This study uses the absolute value of grayscale difference (AVGD) to calculate this. Next, the matching cost aggregation part is calculated. Currently, there are two main types of methods: global cost clustering methods and local cost clustering methods. The former obtains an energy function based on the initial matching cost and then optimizes this function to the global minimum. The latter uses superimposed processing windows to improve the reliability of the matching cost clustering operation. Finally, the stereo vision disparity is obtained and optimized. These are generally divided into two categories: window-based local SMA algorithms and global SMA algorithms. The former selects the optimal point within the window frame after stereo matching cost aggregation. The latter requires first designing an energy evaluation function, and then calculating the minimum energy value according to the optimization method to obtain the best disparity matching, that is, the pixel matching relationship that minimizes the above function.

[0048] As a feasible implementation method, the towing crew controls the camera movement to determine whether the estimated pose error exceeds the set threshold. If so, repositioning is required; otherwise, the subsequent steps can be carried out. Random ferns are used to detect global loops. If there is a global loop, the repositioning tracking algorithm in the previous step is used to estimate the pose change between frames. At the same time, a small number of points are uniformly and randomly extracted, and the node parameters in the graph are optimized by optimizing the constraint equations. Otherwise, the process continues to detect whether there are local loops. If so, the relative pose estimation step is performed, constraints are established, and node parameters are optimized. Finally, the transformed pose is obtained, that is, a three-dimensional damage transformation point map corresponding to the entire vehicle damage image set is constructed.

[0049] As a feasible implementation method, in data collection, a visual sensor is used: Intel RealSense D455 binocular depth camera (global shutter, resolution 1280×720@30fps, depth range 0.5-6m, FOV 86°×57°) and an inertial measurement unit (IMU): built-in 6-axis IMU (accelerometer ±4g, gyroscope ±1000dps). The staff walked around the accident vehicle at a constant speed (speed ≤0.5m / s), keeping the camera 1.2-1.8m away from the vehicle surface. The entire surface of the vehicle was covered in a spiral path (pitch angle ±30°, yaw angle 360° continuous rotation). The acquisition time was 120-180 seconds, and approximately 3600-5400 frames of RGB-D images were obtained. Distortion correction was then performed using a non-local means filter (parameters h = 7, search_window = 21) and the Brown-Conrady model (k1 = -0.15, k2 = 0.03, p1 = p2 = 0). The SLAM algorithm was also used. 1) ORB feature extraction: 1000 feature points were extracted per frame (scaling pyramid with 8 levels and a scaling factor of 1.2).

[0050] 2) Pose estimation: based on PnP+RANSAC (iterations 500, reprojection error threshold 2.5 pixels).

[0051] 3) Closed-loop detection: DBoW2 bag-of-words model (ORB vocabulary tree 10^6 nodes).

[0052] As a feasible implementation, semantic segmentation is first performed for shape feature extraction using DeepLabV3+ (ResNet101 backbone, pre-trained Cityscapes model). A vehicle mask (interference over union ≥ 0.85) is then extracted and model reconstruction is performed using the Canny operator (double threshold 50 / 150). Specifically, this is done by fusion based on the truncated signed distance function (TSDF), combining a voxel resolution of 5mm×5mm×5mm and a cutoff distance of δ=15cm. The input is a binocular camera image of the damaged area (resolution 1280×720, baseline 75mm). Cost calculation is performed using the Census transform (window 9×9). Cost aggregation is then performed using SGM (semi-global matching), with parameters P1=10, P2=120 (penalty coefficient), and eight path directions. Disparity optimization is then performed using sub-pixel fitting (quadratic curve interpolation). Finally, a matching cost contour curve is fitted, and the disparity-cost function and polynomial fit are calculated for each pixel, ultimately extracting the minimum point. Finally, the three-dimensional damage transformation point map is constructed using the damage transformation point map: a sparse 3D point set (density ≥ 200 points / ㎡) and damage quantification indicators: maximum dent depth and deformation volume (accuracy ±2mm).

[0053] Furthermore, based on the coordinate position of the same damage area in the whole vehicle damage image set, the three-dimensional image of the damage transformation point is structurally fused with the initial three-dimensional structural model to obtain a real-time three-dimensional structural model.

[0054] Furthermore, the vehicle database is used to query the vehicle model information corresponding to the accident vehicle, and based on the vehicle model information, the corresponding standard vehicle model 3D model is obtained.

[0055] Furthermore, the real-time 3D architecture model is compared to the 3D model of the standard vehicle model at the same scale to determine the volume of the non-overlapping area. Finally, based on the ratio of the non-overlapping area volume to the 3D model of the standard vehicle model, the damage structure of the accident vehicle is manually graded to determine the loading risk level. The loading risk levels include normal loading risk level, low-risk loading risk level, and high-risk loading risk level.

[0056] In one embodiment of the present invention, the loading risk level of the accident vehicle can also be obtained by comparing the real-time three-dimensional structural model of the accident vehicle with the three-dimensional model of the standard vehicle model to determine the loading level under the damaged structure.

[0057] In one embodiment, Figure 2 This is a schematic diagram of the structure of a three-dimensional model of a standard vehicle model provided in an embodiment of the present application, such as Figure 2As shown in the figure, based on the coordinate location of the same damaged area in the vehicle damage image set, the 3D image of the damage transformation points is structurally fused with the initial 3D structural model to obtain a real-time 3D structural model. Specifically, the initial 3D structural model is deformed and fused to the corresponding area, resulting in a real-time 3D structural model that reflects the damage to the accident vehicle. The vehicle database is then queried for information on the vehicle model corresponding to the accident vehicle. Based on this information, the corresponding 3D model of the standard vehicle model is obtained. The volume of the non-overlapping area is then determined based on a comparison of the same scale and calculation of the non-overlapping area volume. Finally, the proportion of the non-overlapping area volume to the standard vehicle model 3D model is used to determine the loading level of the accident vehicle under the damaged structure, thereby determining the loading risk level (normal, light risk, or high risk).

[0058] S102. If the loading risk level is manually determined to be a high-risk loading risk level, a deformable convolutional network is used to identify damage characteristics at multiple scales on a set of vehicle damage images taken from multiple angles of the accident vehicle, and to determine key damage structure data in each damaged area.

[0059] Specifically, if the loading risk level is manually determined to be high, the collected vehicle damage image set is passed to the deformable convolutional network. The vehicle damage image set is a collection of images from several angles corresponding to each damaged area.

[0060] Furthermore, it is necessary to use a deformable convolutional network to adjust the images in the damage image set under each damage area to a uniform size to obtain damage images of the same size.

[0061] Furthermore, the pre-acquired vehicle accident damage images are processed to capture key points related to the damage type and determine the trend of line changes in the image. Based on this trend, the fuzzy features in the damage images of the same size are robustly enhanced under the relevant pattern changes, resulting in a data-enhanced damage image.

[0062] Furthermore, the data-enhanced damage image in the vehicle damage image is subjected to type recognition calculations based on the vehicle structural damage deformation characteristics to determine the damage characteristic data of each damaged area. The damage characteristic data includes both visible damage characteristic data and hidden damage characteristic data.

[0063] In one embodiment, damage detection for a full vehicle damage image set utilizes Deformable Convolutional Networks (DCNs) as the computational model backbone, incorporating multiple optimization strategies such as multi-scale training, ALBU data augmentation, focal loss, and soft NMS. This approach ultimately achieves accurate identification of key structures within the damaged areas of accident vehicles. The DCN can be trained by inputting preprocessed images. Compared to traditional convolution, DCNs can adaptively adjust the size and shape of its convolution kernel and receptive field, enabling more flexible capture of complex object shapes, such as irregular vehicle damage. Soft non-maximum suppression can address situations where multiple damage targets overlap in the same area. This is employed during model testing to reduce the risk of falsely discarding overlapping targets, thereby improving detection accuracy and recall. The calculation of the multi-task focal loss function requires comparing the model's predictions with the ground-truth labels during training. Therefore, a hybrid loss function, comprising classification loss, bounding box loss, and mask loss, is employed to evaluate the accuracy of model predictions from multiple perspectives.

[0064] As a feasible implementation method, the deformable convolution kernel in the deformable convolutional network is used to perform feature capture processing related to the dynamic convolution kernel and dynamic receptive field on the data-enhanced damage image in the whole vehicle damage image. Based on the irregular vehicle damage shape template, the data-enhanced damage image is subjected to secondary frame selection feature capture to obtain intuitive damage characteristic data. Among them, the intuitive damage characteristic data is the deformation data of damage to a single vehicle structure. Then, based on the soft non-maximum suppression frame selection algorithm, the fractional boundary overlapping boxes in the data-enhanced damage image are processed with high and low scores. Based on the labeling score of each fractional boundary overlapping box, all bounding boxes are hierarchically captured to obtain hidden damage characteristic data. Among them, the hidden damage characteristic data is the overlapping extrusion deformation data under the multi-structure association of the vehicle.

[0065] In one embodiment, in a deformable convolution, the receptive field can dynamically adjust its size and shape based on the characteristics of the input data, thereby more flexibly adapting to different image features. This deformability enables the convolution window to adaptively deform, more accurately and effectively covering the morphology of the target, especially irregularly shaped targets. That is, the data-enhanced damage image is subjected to feature capture processing related to the dynamic convolution kernel and the dynamic receptive field, completing the analysis of the significant target detection results and identifying the presence of many irregularly shaped targets in vehicle damage. By dynamically adjusting the convolution kernel, the deformable convolution network can more effectively capture and learn the morphological features of these irregular targets to improve the accuracy of damage identification and segmentation, thereby performing feature capture processing on the deformation data of single-structure damage of the vehicle, that is, obtaining intuitive damage characteristic data.

[0066] In one embodiment, the soft non-maximum suppression box selection algorithm is an improved bounding box screening algorithm. Unlike traditional non-maximum suppression, soft non-maximum suppression does not completely remove other boxes that overlap with high-scoring bounding boxes, but rather reduces the scores of these boxes based on the degree of overlap. This is particularly important for vehicle damage identification tasks because, during the data collection and annotation process, it can be observed that different vehicle damage categories may have overlapping positions. Soft non-maximum suppression can reduce the risk of mis-discarding overlapping targets, thereby improving detection accuracy and recall. That is, the overlapping extrusion deformation data under the multi-structure association of the vehicle is subjected to hierarchical feature capture under structural overlap, and ultimately the hidden damage characteristic data is obtained, thereby completing the identification of the main types of core damage characteristic structures in the accident vehicle damage area.

[0067] Furthermore, a preset hybrid loss function is used to evaluate the focal type of key damage structures within each damage region's damage characteristic data. Based on the location, segmentation, and classification of the damage targets, the key damage structure data within each damage region is identified and determined. Key damage structures include dents, cracks, wrinkles, perforations, and structural loss. The hybrid loss function comprises classification loss, bounding box loss, and mask loss. Specifically, to alleviate the imbalance in the difficulty of classifying various types of damage in image datasets, a focal loss is introduced within the damage classification subtask. The core idea of ​​the focal loss is to deemphasize easy-to-classify samples and focus on difficult-to-classify ones, assigning greater weight to samples that are misclassified or have high uncertainty. The base classification loss, bounding box loss, and mask loss tasks then evaluate the focal type of key damage structures within each damage region's damage characteristic data. Ultimately, the instance segmentation task is completed, identifying and determining the key damage structure data within each damage region.

[0068] S103. Perform correlation processing on the key damaged structural data in each damaged area based on the factors influencing the damage characteristics, and evaluate and obtain the structural safety status data of each damaged area.

[0069] Specifically, the damage types of the key damage structure data in each damage region are first labeled one by one, and the numerical information corresponding to each damage type is recorded. Then, the contribution value of each damage type and the corresponding numerical information in the current damage region is calculated to obtain the factor contribution value of each damage type in each damage region.

[0070] It should be noted that SHAP (Shapley Additive Explanation, SHAP) is an algorithm based on cooperative game theory for explaining the impact of factors on model predictions. It approximates the output of the prediction model as the sum of the contributions of each input factor, and each feature is associated with a contribution value.

[0071] Furthermore, the SHAP algorithm is used to calculate the marginal contribution weighted average of the contribution value of each factor in each damage area under the feature combination, and the marginal contribution weighted average based on the contribution value of each factor is obtained.

[0072] Furthermore, it is necessary to perform an interactive correlation calculation on the weighted average of the marginal contributions corresponding to each damage type in the same damage region to obtain the SHAP interaction value. Based on the SHAP interaction value between any two damage types, each damage type is evenly divided to obtain the SHAP value of each damage type in each damage region. In other words, the SHAP algorithm is used to calculate the weighted average of the marginal contributions, i.e., the SHAP value after considering the synergistic effect of the feature combination. The SHAP value is the weighted average of the marginal contributions of each feature in all possible feature combinations.

[0073] Furthermore, a structural safety assessment prediction model must be established. This model is used to comprehensively evaluate key damaged structural data. This model can be constructed using expert scoring and predictive estimation to create a score-based prediction model.

[0074] Furthermore, the SHAP values ​​for each damage type within each damage region were summed to obtain the total SHAP value for each damage region. Interaction analysis of the factor nodes within the total SHAP value was then performed to determine the factor interaction relationships between each damage type. These factor interaction relationships represent the interaction relationships between different damage types.

[0075] Furthermore, using the structural safety assessment prediction model, the total SHAP value in each damage zone is calculated by summing the contributions of the model's fractional prediction values ​​to obtain the percentage of accident damage factors in each damage zone. The percentage of accident damage factors is positively correlated with the total SHAP value. Finally, the percentage of accident damage factors is converted into a fractional value to obtain the structural safety status data for each damage zone. A higher structural safety status data corresponds to a more severe damage level.

[0076] In one embodiment, the key damage structure data of each damaged area obtained above, including damage characteristics and related factors, can be combined with machine learning algorithms (such as random forests, gradient boosting trees, etc.) to train a structural safety assessment prediction model. The model can predict the structural safety status data based on the characteristics. After that, the SHAP algorithm can be applied to the data of each damaged area, that is, the contribution of each factor to the structural safety status data is calculated. The total SHAP value of each damaged area is then converted into a percentage to represent the contribution of the accident damage factor in the area. The percentage of the accident damage factor is numerically converted to obtain the structural safety status data of each damaged area. Finally, the degree of damage of each damaged area is evaluated based on the score in the structural safety status data. The above data can also be visualized for towing staff to assist and facilitate the subsequent manual fixation of the towing staff.

[0077] S104: Based on the real-time three-dimensional structural model, a fixed force balance calculation is performed on the accident vehicle to obtain initial vehicle fixed position data.

[0078] Specifically, the real-time 3D architecture model is first subjected to a body posture angle balance calculation to obtain body posture angle data. Next, a static balance calculation of single-point load-bearing capacity is performed on the real-time 3D architecture model to obtain body single-point load-bearing capacity data. Finally, an automatic compensation calculation for body center of gravity offset is performed on the real-time 3D architecture model to obtain body center of gravity balance data.

[0079] Furthermore, the vehicle body posture angle data, vehicle body single-point bearing capacity data and vehicle body center of gravity balance data obtained above are respectively determined as total constraint conditions, that is, the total constraint conditions of the objective function need to be calculated in advance, which is conducive to the subsequent nonlinear calculation under fixed force of the whole vehicle.

[0080] Furthermore, combined with the objective function corresponding to the overall constraint condition, the nonlinear force balance calculation of the vehicle body fixed points is performed on the real-time 3D architecture model. The stress concentration factor, displacement compensation amount, and number of support points are comprehensively evaluated based on the fitness function to generate the initial vehicle fixed position data of the real-time 3D architecture model.

[0081] In one embodiment, the following steps may be used to obtain the initial vehicle fixed position data:

[0082] (1) First, establish a multi-constraint optimization model: First, determine the total constraint conditions: body posture angle data, vehicle body single point load-bearing data, and vehicle body center of gravity balance data are determined as the total constraint conditions. Then construct the objective function.

[0083] (2) An improved particle swarm algorithm is used to solve the problem: 200 particles are initialized, each representing a set of fixed point coordinates (x, y, z). Then, the stress concentration factor, displacement compensation, and number of support points are comprehensively evaluated based on the fitness function. Finally, a simulated annealing mechanism is introduced. When the number of iterations exceeds 500, the inferior solution is accepted with a probability of 0.95^k.

[0084] (3) Calculation of initial vehicle fixed position data: Based on the above steps, the initial fixed force balance calculation is performed on the real-time three-dimensional structural model, that is, the multi-constraint optimization model and the improved particle swarm algorithm are used to complete the nonlinear force balance analysis of the accident vehicle body in the real-time three-dimensional structural model, thereby obtaining an initial vehicle fixed position data under a relatively general analysis, that is, a simple analysis of the vehicle fixed position data without considering the structural safety data of each damaged area, which includes multiple initial fixed position points.

[0085] S105 , using the structural safety data of each damaged area, the initial vehicle fixed position data is corrected for the local area based on the entire vehicle body to determine the reference vehicle fixed position data.

[0086] Specifically, the structural safety data is first processed using a dynamic classification strategy to obtain safety level areas. This means that the structural safety data is dynamically segmented based on values ​​that do not pass the threshold. Safety level areas include red, yellow, and green levels.

[0087] Furthermore, the real-time 3D architecture model is discretized for the entire vehicle, and a stiffness transfer function between nodes is established. Based on the node positions corresponding to the safety level zones, the corresponding nodes in the stiffness transfer function are labeled to obtain labeled mesh nodes. The labeled mesh nodes have three fixed meanings, corresponding to the colors of the safety level zones.

[0088] Furthermore, the initial vehicle fixed position data is loaded first. A global grid node neighborhood scan is then performed for each initial point in the initial vehicle fixed position data. Based on the node avoidance properties of the locally labeled grid nodes, the corresponding neighboring nodes are subjected to a spherical correction domain node Euclidean distance calculation to obtain the position offset vector for each initial point. The node avoidance properties include: all avoidance properties for red nodes and displacement compensation avoidance properties for yellow nodes. In other words, certain nodes need to be avoided or displacement compensated.

[0089] Furthermore, the position offset vector is used to offset all initial fixed points in the initial vehicle fixed position data to obtain corrected reference vehicle fixed position data. Based on the reference vehicle fixed position data, a three-dimensional visual representation of the accident vehicle's fixed positions is generated. This 3D visual representation of the fixed positions assists towing crews in securing the accident vehicle using the calculated fixed positions.

[0090] In one embodiment, Figure 3 A schematic top view of a trailer flatbed provided in an embodiment of the present application at an optimal fixed position, as shown in FIG. Figure 3 As shown, the process of determining the fixed position data of the reference vehicle mainly includes:

[0091] Step 1: Multi-dimensional calculation:

[0092] (1) Establish a damage area safety assessment matrix: define a five-dimensional evaluation vector S = [K_d, ΔK, μ, A_loss, δ_max], which corresponds to the quantitative parameters of the five damage types (depression damage, cracking damage, wrinkling damage, perforation damage, and structural loss damage), where K_d represents depression damage, ΔK represents perforation damage, μ represents cracking damage, A_loss represents the minimum wrinkling damage, and δ_max represents the maximum cracking damage. Then, the analytic hierarchy process (AHP) is used to determine the weight coefficient: ω = [0.25, 0.30, 0.15, 0.20, 0.10] (calibrated and verified by experts); then calculate the regional safety score: Score_i = Σ(ω_j × S_ij) / Σω_j, normalized to a 0-100 point scale, where i and j represent the damage types present in each area, and S is the number of combined damage types.

[0093] (2) Dynamic grading strategy: Red zone (Score < 40): Direct fixation is prohibited, or cross-region support frames can be installed. Yellow zone (40 ≤ Score < 70): Fixation is allowed but displacement compensation is required. Green zone (Score ≥ 70): Direct fixation is allowed. In other words, the dynamic grading strategy calculation under the node avoidance attribute is completed.

[0094] Step 2: Global grid correction processing:

[0095] (a) Constructing a global stress transfer model for the vehicle body: Discretize the entire vehicle into a NURBS surface mesh (basic resolution 10mm×10mm). Then, establish the stiffness transfer function between nodes.

[0096] (b) Execute the local correction algorithm: First, load the initial fixed point set P_initial = {p_1, p_2, ..., p_n}. Then, for each point p_k∈P_initial, perform a neighborhood scan: a spherical correction domain with a radius R = 300 mm is established with p_k as the center. The safety scores Score_m of all mesh nodes within the domain are then extracted. The position offset vector is then calculated. Finally, the corrected position set P_corrected = {p_k + Δp_k} is generated.

[0097] (c) The coordinate position data of the corrected position set is analyzed and marked in the corresponding position of the real-time three-dimensional architecture model, thereby forming the reference vehicle fixed position data of the current accident vehicle.

[0098] S106: The trailer fixing structure feedback data corresponding to the reference vehicle fixing position data may also be periodically monitored to generate loading and driving safety data. That is, during the transportation process, the safety and stability of the fixed accident vehicle may be continuously monitored.

[0099] Specifically, first, the accident vehicle is loaded and fixed onto the trailer based on the reference vehicle fixing position data, wherein the trailer fixing structure includes at least: a binding fixing structure, an auxiliary fixing buckle and a vehicle limiter.

[0100] Furthermore, a pressure sensor is placed at each fixed point in the reference vehicle fixed position data. The pressure sensor is used to periodically monitor the pressure data of the trailer fixed structure and provide feedback data of the trailer fixed structure.

[0101] Furthermore, the trailer's fixed structure feedback data is tabulated and statistically processed to generate data for monitoring the loading and driving safety of the accident vehicle during transportation. This loading and driving safety data can be periodically fed back to the trailer operator's mobile device to complete the safety monitoring of the accident vehicle during transportation, preventing problems such as stress fatigue or loosening of the straps or fixed structure equipment during transportation.

[0102] In addition, the embodiment of the present application also provides an intelligent fixing device for accident vehicles based on the towing service, such as Figure 4 As shown, the accident vehicle intelligent fixing device 400 based on the towing service specifically includes:

[0103] At least one processor 401. And a memory 402 in communication with the at least one processor 401. The memory 402 stores instructions that can be executed by the at least one processor 401, so that the at least one processor 401 can execute:

[0104] The real-time 3D structural model of the accident vehicle and the 3D model of the standard vehicle are used to determine the loading level under the relevant damaged structure to obtain the loading risk level of the accident vehicle;

[0105] If the loading risk level is high, a deformable convolutional network is used to identify damage characteristics at multiple scales on the vehicle damage image set taken from multiple angles of the accident vehicle, and to determine the key damage structure data in each damaged area.

[0106] The key damaged structural data in each damaged area are processed based on the contribution of factors affecting the damage characteristics, and the structural safety data of each damaged area are evaluated and obtained;

[0107] Based on the real-time 3D structural model, the fixed force balance calculation of the accident vehicle is performed to obtain the initial vehicle fixed position data;

[0108] The initial vehicle fixed position data is corrected for the local area based on the entire vehicle body using the structural safety data of each damaged area to determine the reference vehicle fixed position data.

[0109] The embodiments of the present application utilize multi-angle imaging and damage characteristic identification to more accurately assess vehicle damage, thereby improving the assessment of vehicle structural safety during the securing process. Furthermore, the loading risk level of the accident vehicle can be determined. In particular, for vehicles with high loading risk levels, more stringent securing measures can be implemented to reduce the risk of accidents. Deformable convolutional networks can also be used to identify damage characteristics at multiple scales, more accurately determining the key damaged structures within each damaged area, providing a basis for subsequent securing measures. The SHAP algorithm can also be used to evaluate the contribution of factors influencing the key damage structure data within each damaged area, enabling a more comprehensive assessment of the vehicle's structural safety data. Based on a real-time 3D structural model and structural safety data, the optimal vehicle securing position can be calculated to ensure optimal and stable securing results. Periodic monitoring of the trailer securing structure corresponding to the reference vehicle securing position data can also be performed to continuously generate loading and driving safety data, facilitating monitoring and maintenance of vehicle safety status. Damage identification, risk level assessment, and optimal securing position determination can also be automated, reducing manual intervention and improving work efficiency.

[0110] The various embodiments in this application are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0111] The foregoing description describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0112] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the embodiments of the present application may have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included within the scope of the claims of the present application.

Claims

1. An intelligent fixation method for accident vehicles based on towing services, characterized in that: The method comprises: The damage image set of the entire vehicle taken from multiple angles of the accident vehicle is used to identify damage characteristics at multiple scales and determine the key damage structure data in each damage area; Performing correlation processing on the key damaged structural data in each damaged area based on the factors influencing the damage characteristics, and evaluating and obtaining the structural safety status data of each damaged area; Based on the real-time three-dimensional structural model, a fixed force balance calculation is performed on the accident vehicle to obtain initial vehicle fixed position data; By using the structural safety data of each damaged area, the initial vehicle fixing position data is corrected in a local area based on the entire vehicle body to determine the reference vehicle fixing position data, and the reference vehicle fixing position data is displayed in the form of a three-dimensional graph, thereby providing a reference for the fixing method of the accident vehicle.

2. The intelligent fixing method for accident vehicles based on towing services according to claim 1 is characterized in that: The method further includes the step of obtaining the real-time three-dimensional architecture model, and the step specifically includes: The staff uses handheld visual sensor equipment to collect and process continuous image frames of the accident vehicle at all angles to obtain an image set of the accident vehicle and a corresponding visual odometry; wherein the visual odometry is the cumulative pose estimation position and motion trajectory during the image frame collection process; constructing an initial three-dimensional structural model of the accident vehicle based on the visual odometry and the vehicle appearance features in the accident vehicle image set; Performing a stereo matching cost aggregation calculation on the whole vehicle damage image set of the accident vehicle under the left and right visual differences to obtain a fitting matching cost contour curve of the whole vehicle damage image set; Performing global loop detection on the fitted matching cost contour curve under inter-frame pixel positioning changes, and constructing a three-dimensional damage transformation point map corresponding to the vehicle damage image set based on constraint node parameters; Based on the coordinate position of the same damage area in the whole vehicle damage image set, the damage transformation point three-dimensional image and the initial three-dimensional structural model are structurally fused to obtain the real-time three-dimensional structural model.

3. The intelligent fixing method for accident vehicles based on towing services according to claim 1 is characterized in that: The damage image set of the entire vehicle taken from multiple angles of the accident vehicle is used to identify damage characteristics at multiple scales and determine the key damage structure data in each damaged area, including: The collected whole vehicle damage image set of the accident vehicle is passed to the deformable convolutional network; wherein the whole vehicle damage image set is a collection of images from several angles corresponding to each damaged area; Using a deformable convolutional network, the images in the damage image set under each damage area are resized to a uniform size to obtain damage images of the same size; The pre-acquired vehicle accident damage type image is processed to capture key points of the damage type to determine the trend of graphic line changes; and based on the graphic line change trend, the fuzzy features in the damage image of the same size are subjected to robustness enhancement calculation under the relevant graphic changes to obtain a data-enhanced damage image; Performing type recognition calculations on the data-enhanced damage image under the damage deformation characteristics of the vehicle structure to determine damage characteristic data of each damage area; wherein the damage characteristic data includes: intuitive damage characteristic data and hidden damage characteristic data; Through a preset hybrid loss function, the damage characteristic data of each damage area is evaluated for the focus type under the key damage structure, and based on the positioning, segmentation and classification of the damage target, the key damage structure data in each damage area is identified and determined; wherein, the key damage structure data includes: dent damage, crack damage, wrinkle damage, perforation damage and structural loss damage; the hybrid loss function includes: classification loss, bounding box loss and mask loss.

4. The intelligent fixing method for accident vehicles based on towing services according to claim 3 is characterized in that: Performing type recognition calculations on the data-enhanced damage image based on the damage deformation characteristics of the vehicle structure to determine the damage characteristics of each damaged area, specifically including: Performing feature capture processing related to dynamic convolution kernel and dynamic receptive field on the data-enhanced damage image in the whole vehicle damage image, and performing secondary frame selection feature capture on the data-enhanced damage image based on the irregular vehicle damage shape template to obtain the intuitive damage characteristic data; The fractional boundary overlapping boxes in the data enhanced damage image in the whole vehicle damage image are subjected to high and low score marking processing, and based on the marking score of each fractional boundary overlapping box, hierarchical feature capture is performed on all boundary boxes to obtain the hidden damage characteristic data.

5. The intelligent fixing method for accident vehicles based on towing services according to claim 1 is characterized in that: Before performing correlation processing on the key damaged structural data in each damaged area based on the contribution of factors affecting the damage characteristics and evaluating and obtaining the structural safety status data of each damaged area, the method further includes: The marginal contribution weighted average calculation is performed on the contribution value of each factor in each damage area of ​​the vehicle damage image under the feature combination to obtain the marginal contribution weighted average value; The weighted average of the marginal contribution corresponding to each damage type in the same damage area is interactively correlated to obtain the SHAP interaction value. Based on the SHAP interaction value between any two damage types, each damage type is evenly divided to obtain the SHAP value of each damage type in each damage area.

6. The intelligent fixing method for accident vehicles based on towing services according to claim 5 is characterized in that: The key damaged structural data in each damaged area are processed based on the contribution of factors affecting the damage characteristics, and the structural safety data of each damaged area are evaluated and obtained, specifically including: Establishing a structural safety assessment prediction model; wherein the structural safety assessment prediction model is used to perform comprehensive factor assessment processing on the key damaged structure data; The structural safety assessment prediction model is used to calculate the total contribution of the model score prediction value to the total SHAP value in each damage area to obtain the percentage of accident damage factors in each damage area; wherein the percentage of accident damage factors is positively correlated with the total SHAP value; The percentage of the accident damage factor is converted into a fractional value to obtain the structural safety status data of each damaged area; wherein, the higher the score in the structural safety status data, the more serious the damage degree.

7. The intelligent fixing method for accident vehicles based on towing services according to claim 1, characterized in that: Based on the real-time 3D structural model, a fixed force balance calculation is performed on the accident vehicle to obtain initial vehicle fixed position data, specifically including: Performing a vehicle body attitude angle balance calculation on the real-time three-dimensional architecture model to obtain vehicle body attitude angle data; Performing static balance calculation on the single-point load-bearing capacity of the real-time three-dimensional structural model to obtain single-point load-bearing capacity data of the vehicle body; Performing automatic compensation calculation on the real-time three-dimensional architecture model in response to vehicle center of gravity offset to obtain vehicle center of gravity balance data; respectively determining the vehicle body posture angle data, the vehicle body single point bearing capacity data, and the vehicle body center of gravity balance data as total constraint conditions; The nonlinear force balance calculation of the vehicle body fixed points of the real-time three-dimensional architecture model is performed using the objective function corresponding to the total constraint condition to obtain initial vehicle fixed position data of the real-time three-dimensional architecture model.

8. The intelligent fixing method for accident vehicles based on towing services according to claim 1 is characterized in that: The initial vehicle fixed position data is corrected for a local area based on the entire vehicle body using the structural safety data of each damaged area to determine reference vehicle fixed position data, specifically including: Processing the structural safety data using a dynamic grading strategy to obtain safety level areas; wherein the safety level areas include: red level areas, yellow level areas, and green level areas; Discretize the real-time three-dimensional architecture model into a whole vehicle and establish a stiffness transfer function between nodes; According to the node positions corresponding to the safety level areas, corresponding nodes in the inter-node stiffness transfer function are labeled to obtain labeled mesh nodes; wherein the labeled mesh nodes have three fixed meanings and correspond to the colors of the safety level areas; Loading the initial vehicle fixed position data; A global grid node neighborhood scan is performed for each initial point in the initial vehicle fixed position data. Based on the node avoidance attributes of the annotated grid nodes in the local area, the corresponding neighboring nodes are subjected to a node Euclidean distance calculation in a spherical correction domain to obtain a position offset vector for each initial point. The node avoidance attributes include: all avoidance attributes for red nodes and displacement compensation avoidance attributes for yellow nodes. By using the position offset vector, all initial fixed points in the initial vehicle fixed position data are offset to obtain the corrected reference vehicle fixed position data.

9. The intelligent fixing method for accident vehicles based on towing services according to claim 1, characterized in that: After the initial vehicle fixed position data is corrected for a local area based on the entire vehicle body using the structural safety data of each damaged area to determine reference vehicle fixed position data, the method further includes: Based on the reference vehicle fixing position data, the accident vehicle is loaded and fixed onto the trailer; wherein the trailer fixing structure includes at least: a binding fixing structure, an auxiliary fixing buckle and a vehicle limiter; placing a pressure sensor at each fixed point in the reference vehicle fixed position data; Periodically monitoring the pressure data of the trailer fixing structure through the pressure sensor, and feeding back and obtaining the trailer fixing structure feedback data; The feedback data of the trailer fixed structure is statistically processed in tabular form to generate loading and driving safety data for monitoring the accident vehicle during transportation.

10. An intelligent fixing device for accident vehicles based on towing services, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the intelligent fixing method for an accident vehicle based on a towing service according to any one of claims 1 to 9.

Images