A method, device, equipment and medium for constructing a three-dimensional model of an accident vehicle

Abstract

The application provides a construction method, device and equipment of a three-dimensional model of an accident vehicle and a medium, effectively solving the problem that a traditional modeling method cannot accurately restore the three-dimensional structure and damage characteristics of an accident vehicle under the input condition of sparse images. The method comprises the following steps: acquiring multi-view sparse images of an accident vehicle, performing accident vehicle body extraction and damage area identification on the multi-view sparse images to obtain multi-view vehicle body images and damage identification results; based on the multi-view vehicle body images and the damage identification results, multi-view geometric constraint information, structure priors of the accident vehicle and damage constraint information are respectively constructed, and are fused with the vehicle body images to obtain an initial three-dimensional model of the accident vehicle; in view of the defects of the initial three-dimensional model, the initial three-dimensional model is optimized in multiple dimensions to obtain a three-dimensional model of the accident vehicle for application in various application scenarios.

Description

Technical Field

[0001] This application relates to the field of three-dimensional modeling technology, and more specifically, to a method, apparatus, equipment, and medium for constructing a three-dimensional model of an accident vehicle. Background Technology

[0002] In existing technologies, 3D models of accident vehicles typically rely on high-density 360-degree surround view images, laser point clouds, or dedicated 3D scanning equipment. These methods suffer from high acquisition costs, poor on-site adaptability, and stringent requirements for acquisition conditions. In real traffic accident scenarios, images of accident vehicles often originate from on-site investigation photos, surveillance screenshots, law enforcement records, or images captured by mobile devices. These images generally suffer from limited shooting angles, discontinuous angle distributions, severe occlusion, complex background interference, and irregular deformation of the vehicle's damaged areas. Consequently, traditional modeling methods struggle to accurately reconstruct the 3D structure and damage features of accident vehicles.

[0003] Therefore, it is necessary to propose a method that can integrate prior information on vehicle structure and damage constraints under sparse image input conditions to achieve high-fidelity construction of 3D models of accident vehicles, so as to improve the integrity, realism and engineering application value of 3D assets generated from accident vehicles. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method, apparatus, equipment and medium for constructing a three-dimensional model of an accident vehicle, which effectively solves the problem that traditional modeling methods cannot accurately recover the three-dimensional structure and damage features of the accident vehicle under the condition of sparse image as input.

[0005] In a first aspect, embodiments of this application provide a method for constructing a three-dimensional model of an accident vehicle, the method comprising: Acquire multi-view sparse images of the accident vehicle, and perform vehicle body extraction and damage area identification on the multi-view sparse images to obtain multi-view vehicle body images and damage identification results. Based on the multi-view vehicle main body images and damage recognition results, multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information are constructed respectively, and fused with the multi-view vehicle main body images to obtain the initial three-dimensional model of the accident vehicle. To address the shortcomings of the initial 3D model, multi-dimensional optimization is performed on the initial 3D model to obtain a 3D model of the accident vehicle, which can be applied to various application scenarios.

[0006] In conjunction with the first aspect, this application provides a first possible implementation of the first aspect, wherein, based on the multi-view vehicle main image and damage recognition results, multi-view geometric constraint information, structural prior of the accident vehicle, and damage constraint information are respectively constructed, including: Spatial pose relationships are established based on the corresponding features between multi-view images of the main body of the accident vehicle. The multi-view geometric constraint information of the accident vehicle is constructed by using the spatial pose relationship and multi-dimensional geometric constraints.

[0007] In conjunction with the first aspect, this application provides a second possible implementation of the first aspect, wherein establishing spatial pose relationships based on corresponding features between multi-view vehicle body images of the accident vehicle includes: Multiple features are extracted from the vehicle subject images from multiple perspectives, and camera projection relationships are established based on the multiple features between the vehicle subject images from multiple perspectives. By using the corresponding features in the multi-view images and the camera projection relationship, the spatial pose relationship of the vehicle subject image from multiple perspectives is obtained by estimating the camera pose parameters.

[0008] In conjunction with the first aspect, this application provides a third possible implementation of the first aspect, wherein, based on the multi-view vehicle main image and damage recognition results, multi-view geometric constraint information, structural prior of the accident vehicle, and damage constraint information are constructed respectively, and further includes: Based on multiple features between vehicle subject images from multiple perspectives and vehicle scale relationships, the vehicle model template that is closest to the accident vehicle is matched. Based on the damage identification results and vehicle model templates, the structural priors and damage constraint information of the accident vehicle are constructed respectively.

[0009] In conjunction with the first aspect, this application provides a fourth possible implementation of the first aspect, wherein fusing the multi-view vehicle subject images to obtain an initial three-dimensional model of the accident vehicle includes: The multi-view vehicle main body image is used as the main input of the preset 3D generation model, and the multi-view geometric constraint information, structural prior and damage constraint information are used as the conditional input of the 3D generation model. The three-dimensional generation model is controlled to iterate based on the modeling target to obtain the initial three-dimensional model of the accident vehicle corresponding to the main vehicle image.

[0010] In conjunction with the first aspect, this application provides a fifth possible implementation of the first aspect, wherein multi-dimensional optimization of the initial three-dimensional model includes: The defects in the initial 3D model are taken as the regions to be optimized in the initial 3D model, and the optimization method for the regions to be optimized is called. Based on the optimization method, topological repair of the area to be optimized, completion of hidden areas, and texture mapping are performed to obtain a three-dimensional model of the accident vehicle.

[0011] In conjunction with the first aspect, this application provides a sixth possible implementation of the first aspect, wherein performing accident vehicle main body extraction and damage region identification on the multi-view sparse image includes: Remove non-critical regions from the multi-view sparse image and retain critical regions to obtain a multi-view vehicle body image containing only the main body of the accident vehicle. Multi-dimensional damage region identification is performed on vehicle subject images from multiple perspectives to obtain the damage region identification results.

[0012] Secondly, embodiments of this application provide a device for constructing a three-dimensional model of an accident vehicle, the device comprising: The acquisition module is used to acquire multi-view sparse images of the accident vehicle, and to perform accident vehicle main body extraction and damage area identification on the multi-view sparse images to obtain multi-view vehicle main body images and damage identification results. The construction module is used to construct multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information based on the multi-view vehicle main body images and damage recognition results, and fuse them with the multi-view vehicle main body images to obtain an initial three-dimensional model of the accident vehicle. The optimization module is used to optimize the initial 3D model in multiple dimensions to address its defects, thereby obtaining a 3D model of the accident vehicle for application in various scenarios.

[0013] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of any of the methods for constructing a three-dimensional model of an accident vehicle are performed.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of any of the methods for constructing a three-dimensional model of an accident vehicle.

[0015] This application provides a method for constructing a 3D model of an accident vehicle. The method first acquires multi-view sparse images of the accident vehicle, and performs vehicle body extraction and damage region identification on the multi-view sparse images to obtain multi-view vehicle body images and damage identification results. Then, based on the multi-view vehicle body images and damage identification results, multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information are constructed respectively, and fused with the multi-view vehicle body images to obtain an initial 3D model of the accident vehicle. Finally, to address the defects of the initial 3D model, the initial 3D model is optimized in multiple dimensions to obtain a 3D model of the accident vehicle for application in various scenarios. Based on the above methods, this application does not rely heavily on high-precision laser scanning equipment or high-density continuous view acquisition, and has stronger field applicability and a lower data acquisition threshold. It not only solves the problem that traditional modeling methods are unable to accurately recover the three-dimensional structure and damage features of accident vehicles due to the common problems in real traffic accident scenarios such as limited shooting angles, discontinuous view distribution, severe occlusion, complex background interference, and irregular deformation of vehicle damage areas; it also realizes the high-fidelity construction of three-dimensional models of accident vehicles by fusing prior vehicle structure and damage constraint information under sparse image input conditions, thereby improving the integrity, realism, and engineering application value of the generated three-dimensional assets of accident vehicles, and providing a model foundation for accident reconstruction analysis, simulation testing, and dangerous scene reproduction. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a method for constructing a three-dimensional model of an accident vehicle according to an embodiment of this application is shown. Figure 2 This illustration shows a flowchart of obtaining structural prior and damage constraint information according to an embodiment of this application. Figure 3 A schematic diagram illustrating the process of obtaining a three-dimensional model according to an embodiment of this application is shown; Figure 4 This paper shows a structural block diagram of a device for constructing a three-dimensional model of an accident vehicle according to an embodiment of this application; Figure 5 A structural block diagram of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0019] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0021] Existing technologies cannot achieve high-fidelity construction of 3D models of accident vehicles by fusing prior information on vehicle structure and damage constraints under sparse image input conditions. This leads to the problem that traditional modeling methods are unable to accurately recover the 3D structure and damage features of accident vehicles.

[0022] Based on this, this application provides a method, apparatus, device, and medium for constructing a three-dimensional model of an accident vehicle, which will be described below through embodiments.

[0023] Example 1 To facilitate understanding of this embodiment, a method for constructing a three-dimensional model of an accident vehicle disclosed in this application will first be described in detail. For example... Figure 1 The diagram shows a flowchart of a method for constructing a 3D model of an accident vehicle. This application provides a method for constructing a 3D model of an accident vehicle, the method comprising: S101. Obtain multi-view sparse images of the accident vehicle, and perform accident vehicle main body extraction and damage area identification on the multi-view sparse images to obtain multi-view vehicle main body images and damage identification results. S102. Based on the multi-view vehicle main body image and damage recognition results, multi-view geometric constraint information, structural prior of the accident vehicle and damage constraint information are constructed respectively, and fused with the multi-view vehicle main body image to obtain the initial three-dimensional model of the accident vehicle. S103. To address the shortcomings of the initial three-dimensional model, the initial three-dimensional model is optimized in multiple dimensions to obtain a three-dimensional model of the accident vehicle, which can be applied to various application scenarios.

[0024] In step S101, this application first acquires multi-view sparse images of the accident vehicle, wherein the multi-view sparse images are... J=I 1 ,I 2 ,…I N Where N represents the number of sparse images, it is a set of accident vehicle images acquired from the front, rear, left, right, top, or locally damaged areas of the accident vehicle through vehicle-mounted cameras, mobile terminals, law enforcement recording equipment, drones, or other image acquisition devices. This image set can originate from accident scene images, accident investigation images, video frame extraction images, surveillance images, drone images, or vehicle images acquired by other image acquisition devices. Typically, these images cover at least a portion of the front, rear, sides, top, and locally damaged areas of the accident vehicle, thus reflecting the overall shape and local damage status of the accident vehicle. That is, the multi-view sparse images are a set of accident vehicle images with limited viewpoints, discontinuous viewpoint distribution, or partial occlusion. Accident vehicle subject extraction and damage area identification are performed on the multi-view sparse images to obtain multi-view vehicle subject images and damage identification results. Furthermore, for images containing camera parameter information in the multi-view sparse images, their shooting parameters are extracted as initial values ​​for camera intrinsic parameters; for images without valid camera parameter information, a self-calibration method is used to estimate camera intrinsic parameters.

[0025] In a specific implementation of step S101, one embodiment involves: extracting the main body of the accident vehicle and identifying the damaged area from the multi-view sparse image, including: S1011. Remove non-critical regions from the multi-view sparse image and retain critical regions in the multi-view sparse image to obtain a multi-view vehicle body image containing only the main body of the accident vehicle. S1012. Perform multi-dimensional damage region identification on the vehicle body image from multiple perspectives to obtain the damage region identification result.

[0026] In steps S1011-S1012, this application uses a background removal method to remove non-critical regions from the multi-view sparse image. The non-critical regions include irrelevant areas such as roads, guardrails, trees, pedestrians, and other vehicles in the background, as well as background information such as measuring rulers, roads, buildings, pedestrians, vegetation, rescue facilities, and irrelevant vehicles. The key regions in the multi-view sparse image are retained. The multi-view sparse image can also be preprocessed according to actual needs, specifically including brightness correction, contrast enhancement, distortion correction, scale normalization, and image alignment. Scale normalization uniformly scales the long side of the multi-view sparse image to 1600 pixels to ensure that different images have a consistent input scale. Furthermore, the application obtains the main body of the accident vehicle and corresponding multi-view images of the vehicle body through target segmentation methods to improve the stability and accuracy of subsequent feature extraction and 3D model. This application also performs multi-dimensional damage region identification on the multi-view images of the accident vehicle body based on damage detection and segmentation recognition methods, obtaining the damage region identification results. The multi-dimensional damage region identification specifically includes whether there is a collision, whether there is local deformation, whether there are missing parts, whether there is breakage, whether there is dent, and whether cracks have occurred. Therefore, the obtained damage region identification results not only include at least one of the collision deformation region, missing region, broken region, dented region, and crack region, but also damage region information, specifically one or more of the following: damage location, boundary range, local deformation, missing parts, damage type, and damage degree.

[0027] In step S102, after obtaining multi-view vehicle body images and damage recognition results, this application constructs multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information based on the multi-view vehicle body images and damage recognition results. The damage constraint information is further extracted from the damage area recognition results and is used to describe the spatial location, boundary range, local deformation, component missingness, damage type, and damage degree of the damaged parts of the accident vehicle. This information is used to constrain the modeling process of converting multi-view vehicle body images into a 3D generation model for consistency in vehicle shape, component position, and spatial consistency of the damaged area under different viewpoints. After obtaining the multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information, they are fused with the vehicle body images to obtain an initial 3D model of the accident vehicle. This initial 3D model reflects the basic shape and main damage features of the accident vehicle.

[0028] In a specific implementation of step S102, one embodiment is as follows: based on the multi-view vehicle main image and damage recognition results, multi-view geometric constraint information, structural prior of the accident vehicle, and damage constraint information are constructed, including: S10211. Establish spatial pose relationships based on the corresponding features between multi-view vehicle main images of the accident vehicle; S10212. Through the spatial pose relationship and multi-dimensional geometric constraints, the multi-view geometric constraint information of the accident vehicle is constructed.

[0029] In steps S10211-S10212, since the vehicle body images from multiple perspectives all belong to the same accident vehicle, there is a corresponding relationship between their features. Therefore, this application establishes a spatial pose relationship based on the corresponding features extracted from the vehicle body images from multiple perspectives of the accident vehicle. The multi-dimensional geometric constraints set by this application include reprojection consistency constraints, contour consistency constraints, key structural part consistency constraints, and damage area consistency constraints. Based on the spatial pose relationship and the multi-dimensional geometric constraints, the multi-view geometric constraint information of the accident vehicle is constructed and uniformly represented as follows: ; in, This indicates a reprojection consistency constraint. This indicates a contour consistency constraint. This indicates the consistency constraint of projection for key structural components. This indicates the constraints corresponding to the location of the damaged area. , and The weighting coefficients are used to ensure that the projection position of the 3D point in each viewpoint image remains consistent with the original observation position; the contour consistency constraint ensures that the model projection contour matches the vehicle contour in the image; the key structural part consistency constraint ensures that the spatial position of key parts such as doors, headlights, hoods, and tires remains consistent under different viewpoints; the damage area consistency constraint ensures that the correspondence and spatial distribution of the damage area remain consistent under different viewpoints. Through multi-view geometric constraints, the consistency of the overall shape of the accident vehicle, the position of its components, and the damage area in 3D space under different viewpoints can be enhanced.

[0030] In a specific implementation of step S10211, one embodiment is as follows: establishing spatial pose relationships based on corresponding features between multi-view vehicle body images of the accident vehicle, including: A1. Extract various features from the vehicle subject images from multiple perspectives, and establish camera projection relationships based on the various features between the vehicle subject images from multiple perspectives. A2. By estimating the camera pose parameters using the corresponding features in the multi-view images and the camera projection relationship, the spatial pose relationship of the vehicle subject image from multiple perspectives is obtained.

[0031] In steps A1-A2, this application extracts multiple features from the multi-view vehicle subject images, specifically including the vehicle's outer contour features, key structural features, and damaged area features. The outer contour features include one or more of the following: overall vehicle body boundary, roof contour, front contour, rear contour, and side contour. The key structural features include one or more corresponding boundary features, positional features, or shape features of the following: doors, windows, bumpers, fenders, hood, headlights, rearview mirrors, tires, and wheel hubs. The damaged area features include one or more of the following: concave boundaries, wrinkle textures, crack boundaries, fracture boundaries, missing area contours, and deformation directions. A camera projection relationship is established based on the multiple features between the multi-view vehicle subject images, where the camera intrinsic parameter matrix for the i-th view is denoted as... The rotation matrix is The translation vector is Then three-dimensional points Projection point in the i-th image satisfy: ; Next, based on the corresponding feature points in the images from different viewpoints, the camera pose parameters are estimated. These camera pose parameters are solved by minimizing the reprojection error, and the expression is: ; in, Let M represent the projection function, and M represent the number of feature points involved in the matching. After completing the camera pose estimation, the spatial pose relationship of the images from each viewpoint in a unified three-dimensional coordinate system is obtained.

[0032] In the specific implementation of step S102, another embodiment exists as follows: Figure 2 As shown, based on the multi-view vehicle body images and damage recognition results, multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information are constructed, and also include: S10221. Based on multiple features between vehicle subject images from multiple perspectives and vehicle scale relationships, match the vehicle model template that is closest to the accident vehicle. S10222. Based on the damage identification results and the vehicle model template, construct the structural prior and damage constraint information of the accident vehicle respectively.

[0033] In steps S10221-S10222, based on the extracted vehicle contour features, key structural features, and vehicle scale relationships, the vehicle model template closest to the accident vehicle is matched from a preset vehicle template library. This vehicle model template characterizes the overall geometric contour relationship, key component topological connection relationship, body proportion relationship, and relative spatial position relationship of the accident vehicle. The structural prior is then established based on the corresponding vehicle model template. Alternatively, the structural prior can be established based on component topological relationship modeling or fitting of the overall vehicle geometric proportion relationship. The structural prior of the accident vehicle mainly describes the overall geometric contour relationship, key component topological connection relationship, body proportion relationship, left-right symmetry relationship, and relative spatial position relationship of the components that the accident vehicle should satisfy during the modeling process. Especially when the input viewpoint is limited, local areas are invisible, or information is missing, the structural prior can effectively constrain the rationality of the overall structure of the accident vehicle. Then, based on the damage region identification results, damage constraint information is constructed. This damage constraint information is used to characterize abnormal geometric features of the damage region, such as local asymmetric deformation, collapse, wrinkling, fracture, and missing parts. It includes at least the location constraints, extent constraints, boundary constraints, deformation constraints, damage type constraints, component missing part constraints, and damage severity constraints of the damage region. This application unifies the structural priors and damage constraints as follows: ; in, This represents the vehicle structure prior constraints corresponding to the structural priors. This represents the damage constraint terms corresponding to the damage constraint information. and The corresponding weighting coefficients are as follows. The prior structural constraints of the vehicle are used to maintain the rationality of the overall structure of the accident vehicle; the damage deformation constraints are used to preserve the true damage characteristics of the locally damaged parts of the accident vehicle. Based on the above formulas, the prior structural information and damage constraint information of the accident vehicle are obtained.

[0034] In the specific implementation of step S102, another embodiment is as follows: the initial three-dimensional model of the accident vehicle is obtained by fusing the multi-view vehicle main body image, including: S10231. The multi-view vehicle main body image is used as the main input of the preset three-dimensional generation model, and the multi-view geometric constraint information, structural prior and damage constraint information are used as the condition input of the three-dimensional generation model. S10232. Control the three-dimensional generation model to iterate based on the modeling target to obtain the initial three-dimensional model of the accident vehicle corresponding to the vehicle body image.

[0035] In steps S10231-S10232, this application uses the multi-view vehicle main image as the main input of a preset 3D generation model, and the multi-view geometric constraint information, structural prior, and damage constraint information as conditional inputs to the 3D generation model. These are used to input a pre-trained image into the 3D generation model, controlling the 3D generation model to iterate based on the modeling objective to obtain an initial 3D model of the accident vehicle corresponding to the vehicle main image. The multi-view vehicle main image to 3D generation model can be any one of the implicit representation model, point cloud model, and mesh model corresponding to image-to-point cloud, image-to-implicit surface, or image-to-mesh generation networks, respectively, or it can be an explicit surface model. In this embodiment, a meshed initial 3D model is preferably output for subsequent mesh repair and texture mapping. To uniformly describe the initial 3D model generation process, the modeling objective is represented as: E = ; in, Represents multi-view geometric constraint terms. Represents the structural priors and damage constraints. Represents the regularization term. For regularization weights.

[0036] In step S103, after obtaining the initial 3D model of the accident vehicle, this application requires further optimization to improve its integrity and usability. Therefore, the initial 3D model is processed to obtain the final 3D model of the accident vehicle. Addressing the deficiencies of the initial 3D model, this application performs multi-dimensional optimization to obtain the 3D model of the accident vehicle. The 3D model includes at least one or more of the following information: overall 3D geometric shape information of the vehicle, surface texture information, and component semantic label information. Depending on the actual application needs, the 3D model of the accident vehicle can be further semantically annotated and simulated to form 3D model data of the accident vehicle that can be used for accident reconstruction analysis, simulation testing, and dangerous scene reproduction, thus applicable to various application scenarios.

[0037] In the specific implementation of step S103, one embodiment is as follows: Figure 3 As shown, the initial 3D model undergoes multi-dimensional optimization, including: S1031. Take the defects in the initial three-dimensional model as the region to be optimized in the initial three-dimensional model, and call the optimization method for the region to be optimized; S1032. According to the optimization method, perform topological repair of the area to be optimized, completion of hidden areas, and texture mapping to obtain a three-dimensional model of the accident vehicle.

[0038] In step S1031- In S1032, this application identifies defects in the initial 3D model, uses these defects as the areas to be optimized in the initial 3D model, and calls optimization methods for these areas. The types include completion, repair, and mapping. Based on the optimization methods, topological repair, hidden area completion, and texture mapping are performed on the areas to be optimized to obtain a 3D model of the accident vehicle. Specifically: First, the initial 3D model undergoes mesh optimization processing, including removing outliers, repairing holes on the model surface, eliminating local self-intersecting structures, and repairing non-manifold structures to improve the continuity and topological stability of the model mesh. Second, areas in the initial 3D model that are not fully restored due to occlusion or insufficient image acquisition are completed. This completion process combines existing model structures, geometric relationships between adjacent areas, and overall vehicle morphology information to reconstruct invisible areas, improving the overall integrity of the model. Then, the completed 3D model undergoes texture mapping processing, which maps color information, material information, and damage details from the input image to the surface of the 3D model to restore the appearance features of the accident vehicle surface. Furthermore, the 3D model after texture mapping undergoes detail restoration processing. This process includes local enhancement of damaged edges, wrinkled areas, cracked areas, and transition areas of missing regions to improve the model's accuracy in representing the accident damage state. Semantic annotation can also be applied to vehicle component regions and damaged areas in the 3D model to create 3D assets of the accident vehicle with structural and damage attributes.

[0039] Example 2 This application also provides a vehicle active and passive safety calibration device, such as Figure 4 The diagram shows a block diagram of a vehicle active and passive safety calibration device. This device performs functions corresponding to the steps of executing a vehicle active and passive safety calibration method on a terminal device as described above. The device can be understood as a server component including a processor. The device for constructing a 3D model of an accident vehicle as described in this application includes: The acquisition module is used to acquire multi-view sparse images of the accident vehicle, and to perform accident vehicle main body extraction and damage area identification on the multi-view sparse images to obtain multi-view vehicle main body images and damage identification results. The construction module is used to construct multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information based on the multi-view vehicle main body images and damage recognition results, and fuse them with the multi-view vehicle main body images to obtain an initial three-dimensional model of the accident vehicle. The optimization module is used to optimize the initial 3D model in multiple dimensions to address its defects, thereby obtaining a 3D model of the accident vehicle for application in various scenarios.

[0040] In one feasible implementation, the building module includes: Spatial pose relationships are established based on the corresponding features between multi-view images of the main body of the accident vehicle. The multi-view geometric constraint information of the accident vehicle is constructed by using the spatial pose relationship and multi-dimensional geometric constraints.

[0041] In one feasible implementation, the building module further includes: Multiple features are extracted from the vehicle subject images from multiple perspectives, and camera projection relationships are established based on the multiple features between the vehicle subject images from multiple perspectives. By using the corresponding features in the multi-view images and the camera projection relationship, the spatial pose relationship of the vehicle subject image from multiple perspectives is obtained by estimating the camera pose parameters.

[0042] In one feasible implementation, the building module also includes: Based on multiple features between vehicle subject images from multiple perspectives and vehicle scale relationships, the vehicle model template that is closest to the accident vehicle is matched. Based on the damage identification results and vehicle model templates, the structural priors and damage constraint information of the accident vehicle are constructed respectively.

[0043] In one feasible implementation, the building module further includes: The multi-view vehicle main body image is used as the main input of the preset 3D generation model, and the multi-view geometric constraint information, structural prior and damage constraint information are used as the conditional input of the 3D generation model. The three-dimensional generation model is controlled to iterate based on the modeling target to obtain the initial three-dimensional model of the accident vehicle corresponding to the main vehicle image.

[0044] In one feasible implementation, the optimization module includes: The defects in the initial 3D model are taken as the regions to be optimized in the initial 3D model, and the optimization method for the regions to be optimized is called. Based on the optimization method, topological repair of the area to be optimized, completion of hidden areas, and texture mapping are performed to obtain a three-dimensional model of the accident vehicle.

[0045] In one feasible implementation, the acquisition module includes: Remove non-critical regions from the multi-view sparse image and retain critical regions to obtain a multi-view vehicle body image containing only the main body of the accident vehicle. Multi-dimensional damage region identification is performed on vehicle subject images from multiple perspectives to obtain the damage region identification results.

[0046] Example 3 This application also provides an electronic device, such as Figure 5 As shown, it includes: a processor 501, a memory 502, and a bus 503. The memory 502 stores machine-readable instructions that can be executed by the processor 501. When the electronic device is running, the processor 501 and the memory 502 communicate through the bus 503. When the machine-readable instructions are executed by the processor 501, the steps of any one of the methods for constructing a three-dimensional model of an accident vehicle are performed.

[0047] Example 4 This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of any one of the methods for constructing a three-dimensional model of an accident vehicle.

[0048] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0049] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0050] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0051] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a platform server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0052] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for constructing a three-dimensional model of an accident vehicle, characterized in that, The method includes: Acquire multi-view sparse images of the accident vehicle, and perform vehicle body extraction and damage area identification on the multi-view sparse images to obtain multi-view vehicle body images and damage identification results. Based on the multi-view vehicle main body images and damage recognition results, multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information are constructed respectively, and fused with the multi-view vehicle main body images to obtain the initial three-dimensional model of the accident vehicle. To address the shortcomings of the initial 3D model, multi-dimensional optimization is performed on the initial 3D model to obtain a 3D model of the accident vehicle, which can be applied to various application scenarios.

2. The method according to claim 1, characterized in that, Based on the multi-view vehicle main body images and damage recognition results, multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information are constructed, including: Spatial pose relationships are established based on the corresponding features between multi-view images of the main body of the accident vehicle. The multi-view geometric constraint information of the accident vehicle is constructed by using the spatial pose relationship and multi-dimensional geometric constraints.

3. The method according to claim 2, characterized in that, Spatial pose relationships are established based on the corresponding features between multi-view images of the vehicle's main body in the accident, including: Multiple features are extracted from the vehicle subject images from multiple perspectives, and camera projection relationships are established based on the multiple features between the vehicle subject images from multiple perspectives. By using the corresponding features in the multi-view images and the camera projection relationship, the spatial pose relationship of the vehicle subject image from multiple perspectives is obtained by estimating the camera pose parameters.

4. The method according to claim 1, characterized in that, Based on the multi-view vehicle main image and damage recognition results, multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information are constructed, and also include: Based on multiple features between vehicle subject images from multiple perspectives and vehicle scale relationships, the vehicle model template that is closest to the accident vehicle is matched. Based on the damage identification results and vehicle model templates, the structural priors and damage constraint information of the accident vehicle are constructed respectively.

5. The method according to claim 1, characterized in that, The initial 3D model of the accident vehicle is obtained by fusing the multi-view vehicle subject images, including: The multi-view vehicle main body image is used as the main input of the preset 3D generation model, and the multi-view geometric constraint information, structural prior and damage constraint information are used as the conditional input of the 3D generation model. The three-dimensional generation model is controlled to iterate based on the modeling target to obtain the initial three-dimensional model of the accident vehicle corresponding to the main vehicle image.

6. The method according to claim 1, characterized in that, The initial 3D model is optimized in multiple dimensions, including: The defects in the initial 3D model are taken as the regions to be optimized in the initial 3D model, and the optimization method for the regions to be optimized is called. Based on the optimization method, topological repair of the area to be optimized, completion of hidden areas, and texture mapping are performed to obtain a three-dimensional model of the accident vehicle.

7. The method according to claim 1, characterized in that, The process of extracting the main body of the accident vehicle and identifying the damaged area from the multi-view sparse image includes: Remove non-critical regions from the multi-view sparse image and retain critical regions to obtain a multi-view vehicle body image containing only the main body of the accident vehicle. Multi-dimensional damage region identification is performed on vehicle subject images from multiple perspectives to obtain the damage region identification results.

8. A device for constructing a three-dimensional model of an accident vehicle, characterized in that, The device includes: The acquisition module is used to acquire multi-view sparse images of the accident vehicle, and to perform accident vehicle main body extraction and damage area identification on the multi-view sparse images to obtain multi-view vehicle main body images and damage identification results. The construction module is used to construct multi-view geometric constraint information, structural priors of the accident vehicle, and damage constraint information based on the multi-view vehicle main body images and damage recognition results, and fuse them with the multi-view vehicle main body images to obtain an initial three-dimensional model of the accident vehicle. The optimization module is used to optimize the initial 3D model in multiple dimensions to address its defects, thereby obtaining a 3D model of the accident vehicle for application in various scenarios.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of a method for constructing a three-dimensional model of an accident vehicle as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of a method for constructing a three-dimensional model of an accident vehicle as described in any one of claims 1 to 7.

Images