Vehicle frame number one-to-one high-resolution texture reduction method and system and storage medium

Abstract

The invention discloses a vehicle frame number one-to-one high-resolution texture restoration method and system and a storage medium, the method is applied to a trinocular stereo vision camera system, and the trinocular stereo vision camera system comprises a binocular stereo vision camera and an ultra-high-definition color camera. The binocular stereo vision camera comprises two infrared cameras, an infrared structured light projector and a high-definition color camera. The method comprises the following steps: carrying out multi-angle shooting on a frame number by using the binocular stereo vision camera; processing the shot image to generate a frame number equal-ratio front view; when the binocular stereoscopic vision camera is used for shooting the frame number, the ultra-high-definition color camera is used for shooting the frame number so as to obtain a high-resolution color image; according to the color image, obtaining an area image only containing the frame number; and matching the frame number equal-ratio front view with the area-only image, and combining the frame number equal-ratio front view with the area-only image to form a high-resolution ultra-high-definition frame number equal-ratio front view. According to the method, the vehicle frame number size texture can be conveniently restored with high precision.

Description

Technical Field

[0001] This invention relates to the field of image processing and three-dimensional coordinate calculation technology, and in particular to a method, system, terminal device, and computer-readable storage medium for high-resolution texture restoration of vehicle identification number (VIN) at a ratio of 1:1. Background Technology

[0002] Vehicle identification number (VIN) comparison is a common part of vehicle inspections. Currently, the traditional method of copying the VIN onto paper is still used in most cases. The advantage of this method is that it can achieve high-precision restoration of the VIN size and letter outline through physical contact. The disadvantage is that it cannot retain the color and texture information of the VIN. Therefore, some VIN texture restoration technologies have gradually emerged. In the field of VIN size texture restoration, the existing technologies mainly have the following shortcomings: (1) Laser ranging type: It relies on a device with a three-point laser ranging sensor to calculate the plane and distance of the VIN. However, the laser is not visible in the camera and cannot be linked with the camera. After the plane is corrected, it is difficult to accurately know the distance between the plane and the camera. The error is large and it can only remind the user to align the VIN parallel. In general, it is costly, has a poor user experience, lacks strong constraints to ensure accuracy, and has high operational requirements. (2) Ruler-dependent type: It is necessary to magnetically attach or stick a ruler or marker on the surface of the VIN and infer the actual size from the distance of the ruler. However, this operation is cumbersome and can only calculate the coordinates of the ruler plane. The ruler has a thickness, and the thickness of the ruler will inevitably lead to depth error. (3) Mold fixed type scheme: Special molds are required to maintain a fixed distance. However, the molds are large and heavy, and inconvenient to use. (4) Binocular stereo vision system: Traditional schemes consume a lot of energy due to parallax calculation and have limited resolution. Although they can achieve pure vision without rulers and molds, they are difficult to meet the requirements of high-resolution texture restoration.

[0003] Furthermore, the location of the vehicle identification number (VIN) can be tricky, such as under the seat or in a small recess, making it impossible to fit a large ruler, and because it's not a large flat surface, it's difficult to use a three-point laser to measure the plane, let alone fit a large mold for photography. Therefore, there is an urgent need to develop a high-resolution texture reconstruction solution for VIN sizes that requires no auxiliary equipment, is highly portable, and has an accuracy of up to 0.01mm. Summary of the Invention

[0004] In view of this, the present invention provides a method, system, terminal device and computer-readable storage medium for high-resolution texture restoration of vehicle identification number (VIN) at a ratio of 1:1, which can conveniently and accurately restore the VIN size texture.

[0005] The first objective of this invention is to provide a method for high-resolution texture restoration of vehicle identification number (VIN) at a 1:1 ratio.

[0006] The second objective of this invention is to provide a high-resolution texture reproduction system for vehicle identification numbers (VINs) at a 1:1 ratio.

[0007] The third objective of this invention is to provide a terminal device.

[0008] A fourth objective of this invention is to provide a computer-readable storage medium.

[0009] The first objective of this invention can be achieved by adopting the following technical solution: A method for high-resolution texture reconstruction of vehicle identification number (VIN) at a 1:1 ratio is applied to a trinocular stereo vision camera system. The trinocular stereo vision camera system includes a binocular stereo vision camera and an ultra-high-definition color camera. The binocular stereo vision camera includes two infrared cameras, an infrared structured light projector, and a high-definition color camera. The two infrared cameras and the structured light projector work together to acquire depth information of the VIN, and the high-definition color camera provides color information of the VIN. The method includes: The vehicle identification number (VIN) is captured from multiple angles using a binocular stereo camera; the captured images are processed to generate a scaled-down orthographic view of the VIN. While using a binocular stereo vision camera to photograph the vehicle identification number (VIN), an ultra-high-definition color camera is also used to photograph the VIN to obtain a high-resolution color image; based on the color image, an area image containing only the VIN is obtained; wherein, the color image overlaps with the image captured by the binocular stereo vision camera by at least half. Match the VIN scale front view with the area image containing only the VIN and merge them into a high-resolution ultra-high-definition VIN scale front view.

[0010] Furthermore, the step of matching the scaled-to-scale front view of the vehicle identification number (VIN) with an area image containing only the VIN and merging them into a high-resolution, ultra-high-definition scaled-to-scale front view of the VIN includes: Feature point matching pairs are selected from the scaled front view of the vehicle identification number and the area image containing only the vehicle identification number to establish the correspondence between the two images; The perspective matrix is ​​solved using the RANSAC algorithm for feature point matching pairs; the perspective transformation is performed on the region image containing only the vehicle identification number (VIN) using the perspective matrix to make it consistent with the perspective of the VIN's scaled front view; and the color information in the scaled front view of the VIN is then replaced with the color information in the transformed region image. The dimensions of the VIN (Vehicle Identification Number) in a scaled-down front view after the color information has been replaced are adjusted to achieve a one-to-one high-resolution texture reproduction of the VIN.

[0011] Furthermore, the process of processing the captured image to generate a scaled-down front view of the vehicle identification number includes: Generate a color depth map based on the captured images; Based on the pre-calibrated camera intrinsic parameter matrix, each pixel in the color depth map is converted into xyz coordinates in the camera coordinate system to construct 3D point cloud data; Based on the converted color depth map, locate the vehicle identification number area; Based on the 3D point cloud data of the vehicle identification number (VIN) region, the SVD algorithm is used to suppress noise interference and extract the VIN plane normal vector. Calculate the rotation matrix facing the camera based on the normal vector of the vehicle identification number plane; Based on the rotation matrix, the vehicle identification number plane is rotated and corrected to face the camera. Based on the rotated and corrected VIN plane, a two-dimensional texture image is constructed using ray projection to generate a scaled-down front view of the VIN.

[0012] Furthermore, the step of constructing a two-dimensional texture image using light projection based on the rotated and corrected vehicle identification number plane includes: Calculate the width and height of the VIN area based on the VIN area where the rotated and corrected VIN plane is located; Calculate the scaling ratio based on the calculated width and height and the width and height of the target VIN area; Based on the scaling ratio, determine the width and height of the target image, and initialize the target image with a black background; The process iterates through the 3D point cloud data of the VIN region where the VIN plane is located after rotation correction, mapping the x and y coordinates of each 3D point to the corresponding position in the target image according to the scaling factor; if the x and y coordinates exceed the boundary of the target image, they are skipped; if the target pixel already has a value, the number of overlaps is recorded for calculating the overlap rate; using the nearest neighbor interpolation method, the RGB data corresponding to the 3D point cloud of the VIN region are assigned to the corresponding pixels in the target image, gradually constructing a two-dimensional texture image to generate a scaled front view of the VIN.

[0013] Furthermore, a repair mask is used to repair the scaled front view of the generated vehicle frame number to fill in the empty areas.

[0014] Furthermore, the step of using the SVD algorithm to suppress noise interference and extract the normal vector of the VIN plane based on the 3D point cloud data of the VIN region includes: Based on the set depth value filtering criteria, the 3D point cloud data of the vehicle identification number area is filtered to select valid 3D point cloud data. The SVD method is used to decompose the 3D point cloud data of the filtered vehicle identification number area into a product of matrices U, Σ, and Vt. Transpose matrix U to obtain matrix Ut; find the index of the element with the largest absolute value in each row of matrix Ut and arrange them into a column, denoted as max_abs_u_cols; create an arithmetic sequence shift from 0 to the number of rows in matrix Ut minus 1; multiply shift by the number of columns in matrix Ut and add it to each element of max_abs_u_cols to obtain indices; extract elements from matrix Ut based on indices and obtain the signs of these elements, denoted as signs; Multiply each element in signs by the corresponding element in the corresponding column of matrix Vt to obtain the adjusted matrix Vt; extract the last row of data from the adjusted matrix Vt as the normal vector of the chassis number plane.

[0015] Furthermore, generating a color depth map based on the captured image includes: Images of the vehicle identification number (VIN) are simultaneously acquired by two infrared cameras on the left and right sides. The depth information of each pixel in the image is calculated using structured light or binocular parallax principle. While acquiring VIN images simultaneously by the two cameras, images are also acquired by a high-definition color camera. The color information and depth information of each pixel in the image are fused to generate a color depth map.

[0016] The second objective of this invention can be achieved by adopting the following technical solution: A 1:1 high-resolution texture reconstruction system for vehicle identification numbers (VINs) is applied to a trinocular stereo vision camera system. The trinocular stereo vision camera system includes a binocular stereo vision camera and an ultra-high-definition color camera. The binocular stereo vision camera includes two infrared cameras, an infrared structured light projector, and a high-definition color camera. The two infrared cameras and the structured light projector work together to acquire depth information of the VIN, and the high-definition color camera provides color information of the VIN. The 1:1 high-resolution texture reconstruction system for VINs includes: The image processing module is used to capture the vehicle identification number (VIN) from multiple angles using a binocular stereo vision camera; and to process the captured images to generate a scaled-down orthographic view of the VIN. The image acquisition module is used to simultaneously capture the vehicle identification number (VIN) using a binocular stereo vision camera and an ultra-high-definition color camera to obtain a high-resolution color image; based on the color image, a region image containing only the VIN is obtained; wherein the color image and the image captured by the binocular stereo vision camera overlap by at least half. The image matching and merging module is used to match the full-scale front view of the vehicle identification number (VIN) with an area image containing only the VIN, and merge them into a high-resolution, ultra-high-definition full-scale front view of the VIN.

[0017] The third objective of this invention can be achieved by adopting the following technical solution: A terminal device includes a processor and a memory for storing a processor-executable program. When the processor executes the program stored in the memory, it implements the above-described method for high-resolution texture restoration of vehicle identification number (VIN) at a ratio of 1:1.

[0018] The fourth objective of this invention can be achieved by adopting the following technical solution: A computer-readable storage medium storing a program that, when executed by a processor, implements the above-described method for high-resolution texture reconstruction of vehicle identification numbers (VINs) at a 1:1 ratio.

[0019] The present invention has the following advantages over the prior art: (1) This invention is the first to construct a tri-lens pure vision composite system, which deeply integrates an ultra-high-definition color camera and a binocular stereo vision camera: the binocular stereo vision camera realizes the acquisition of stereo depth information through an infrared camera and a structured light projector, eliminating the dependence of traditional solutions on external auxiliary equipment such as laser rangefinders, rulers, and molds; the ultra-high-definition color camera uses its high pixel characteristics to acquire high-resolution color textures. The two work together to simultaneously complete the calculation of three-dimensional coordinates and the acquisition of high-resolution color information without physical contact or additional devices, constructing a new technical path of "pure vision + high precision", fundamentally solving the common problems of existing solutions such as cumbersome operation, bulky equipment, and dependence on external auxiliary tools, and making a key breakthrough in the field of vehicle identification number texture restoration from "physical contact" to "visual intelligence".

[0020] (2) Through an innovative combination of SVD plane fitting, rotation matrix correction, and ray projection algorithms, a dimensional restoration accuracy of 0.01mm-level error was achieved, far exceeding existing technologies such as laser ranging (large error) and ruler-dependent (affected by thickness error), achieving dimensional accuracy equivalent to manual rubbing, while retaining photo-quality color texture, filling the gap in color information that traditional rubbing cannot restore. In terms of adaptability, for the common tricky locations of VINs such as narrow, recessed, and non-planar areas, the system can complete the inspection without large equipment due to the lightweight characteristics of the pure vision solution. It solves the application bottleneck of traditional mold-fixed and binocular stereo vision system solutions in complex scenarios, becoming the first VIN detection solution that can achieve high-precision restoration in all scenarios, providing core technical support for the automation of vehicle inspection. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a method for high-resolution texture restoration of vehicle identification number (VIN) according to an embodiment of the present invention.

[0023] Figure 2 This is a structural block diagram of a terminal device according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be understood that the specific embodiments described are merely used to explain this application and are not intended to limit this application.

[0025] Example: like Figure 1 As shown, this embodiment provides a method for high-resolution texture reconstruction of vehicle identification numbers (VINs) at a 1:1 ratio, applied to a trinocular stereo vision camera system. This system includes a binocular stereo vision camera and an ultra-high-definition color camera with extremely high image quality. The binocular stereo vision camera comprises two infrared cameras (left and right), an infrared structured light projector, and a high-definition color camera. The two infrared cameras and the structured light projector acquire stereo depth information, while the color camera provides color information. Although this binocular stereo vision camera has three cameras, it is conventionally referred to as a binocular camera. The method includes the following steps: S101. Use a binocular stereo camera to take pictures of the vehicle identification number (VIN) from multiple angles; process the pictures to generate a scaled-down front view of the VIN.

[0026] A high-precision binocular stereo vision camera is fixed to a portable terminal, ensuring that the camera and the terminal's built-in ultra-high-definition color camera are as close as possible, oriented in the same direction, and with more than half of their images overlapping. Multiple angle shots of the vehicle identification number (VIN) are then taken. The equipment must be kept stable during shooting to avoid image blurring due to shaking. Utilizing binocular vision principles, the XYZ coordinates of the VIN in the real world are acquired during shooting, generating a VIN image with depth data. Finally, a scaled-down frontal view of the VIN is output (after scaling it proportionally before printing, it achieves near-perfect consistency in texture and size with the actual VIN plane).

[0027] Furthermore, step S101 specifically includes: (1) Generate a color depth map based on the captured image.

[0028] Images are simultaneously acquired by the left and right cameras of a binocular vision camera. The depth information of each pixel is calculated using structured light or binocular parallax principles. While the left and right cameras are acquiring images simultaneously, a high-definition color camera is also capturing images. The color and depth information of each pixel in the image are fused to generate an RGBD image (color depth map); where RGB refers to the red, green, and blue color images; and D stands for Depth, referring to the depth information of each pixel in the image.

[0029] (2) Convert each pixel in the color depth map to xyz coordinates to construct 3D point cloud data.

[0030] Based on the pre-calibrated camera intrinsic parameter matrix, the pixel coordinates and depth values ​​of each pixel in the RGBD image are converted into xyz coordinates in the real world to construct the 3D point cloud data of the vehicle identification number area.

[0031] (3) Locate the vehicle identification number area based on the converted color depth map.

[0032] The converted RGBD image is input into the trained VIN image localization AI model. The target detection algorithm is used for feature extraction and analysis to automatically select the area where the VIN is located. The image is then cropped to remove irrelevant background information, resulting in an image containing only the VIN.

[0033] The model network structure of the vehicle identification number (VIN) image localization AI model is an existing one. For example, the VIN image localization AI model is formed by training a large number of VIN image samples with models such as YOLO / RT-DETR. It can accurately identify the location of the VIN in the image.

[0034] (4) Based on the 3D point cloud data of the vehicle frame number area, the SVD algorithm is used to suppress noise interference and extract the normal vector of the vehicle frame number plane.

[0035] In RGBD images acquired using a binocular vision camera, due to factors such as binocular matching errors and surface reflections, some pixels may contain invalid data with a depth value of 0. Therefore, it is necessary to preprocess the point cloud data within the vehicle identification number (VIN) location area. By setting depth value filtering conditions, invalid point clouds with a depth of 0 are filtered out, retaining only data with valid depth information.

[0036] Even after filtering, the VIN planar 3D point cloud data still suffers from noise and jitter due to camera system errors. Directly applying the least squares normal equations for planar fitting would fail to yield the optimal solution due to this noise interference. Therefore, a more robust SVD (Singular Value Decomposition) method is employed to decompose the processed 3D point cloud data into a product of three matrices (U / Σ / Vt). This method effectively suppresses noise interference and yields a more robust planar fitting result through mathematical transformation. Then, matrix U is transposed to obtain matrix Ut. Next, the indices of the elements with the largest absolute values ​​in each row of matrix Ut are identified and arranged into columns, denoted as max_abs_u_cols. Then, an arithmetic sequence shift is created from 0 to the row number minus 1 in matrix Ut. After multiplying shift by the column number of matrix Ut, each element of max_abs_u_cols is added to the transformed shift to obtain a new set of indices. Elements are extracted from matrix Ut based on indices, and their signs are obtained, denoted as signs. The sign of each column of matrix Vt is adjusted using `signs`, which means that each element of `signs` is multiplied by the corresponding element of the column in Vt. Finally, the last row of data is extracted from the adjusted matrix Vt, which yields the key geometric parameter used for subsequent texture reconstruction: the chassis number plane normal vector.

[0037] (5) Calculate the rotation matrix facing the camera based on the normal vector of the chassis number plane.

[0038] First, the input vehicle identification number plane normal vector and the camera facing direction vector (x=0, y=0, z=1) are normalized to a magnitude of 1, resulting in two unit vectors. This step ensures that subsequent calculations are not affected by vector length, making the results depend only on vector direction.

[0039] Next, perform a cross product on the two unit vectors to obtain a vector v with a length of 3; at the same time, calculate the dot product of the two unit vectors to obtain a scalar value c; and then calculate the magnitude s of the cross product vector.

[0040] Then, a 3x3 antisymmetric matrix skew_v is constructed based on the vector v, in the following form:

[0041] Finally, combining the identity matrix, the antisymmetric matrix, the dot product, and the cross product, the rotation matrix is ​​calculated using specific mathematical operations. Specifically, the 3x3 identity matrix is ​​added to the antisymmetric matrix skew_v, and then (skew_v@skew_v)*(1-c) / (s*s+eps) (where eps is a minimal value to avoid a denominator of 0). Finally, the resulting matrix is ​​transposed to obtain the rotation matrix required to rotate the chassis number plane to face the camera.

[0042] (6) Based on the rotation matrix, the vehicle frame number plane is rotated and corrected so that it faces the camera.

[0043] By performing a matrix multiplication on the effective portion of the depth information in the 3D point cloud data of the VIN region and the rotation matrix required to rotate the VIN plane to face the camera, the tilted VIN plane can be rotated to face the camera, resulting in point cloud data of the VIN plane facing the camera. This provides standard viewpoint data for subsequent texture reconstruction.

[0044] (7) Based on the rotated and corrected chassis number plane, a two-dimensional texture image is constructed using ray projection to obtain a scaled front view of the chassis number.

[0045] Ray casting: After completing the rotation correction and size adjustment of the VIN 3D point cloud, it needs to be converted into a 2D image using a ray casting algorithm to lay the foundation for subsequent 1:1 scale texture restoration. Since the VIN planar point cloud has already been rotated and corrected in the previous step, each point cloud is scaled proportionally and projected into the 2D image during the ray casting operation. Therefore, the distance between every two horizontally and vertically adjacent pixels in this 2D image is the same, which achieves a 1:1 scale restoration of the VIN texture information. In terms of size, it is equivalent to traditional manual photocopying and retains photo-quality color texture information.

[0046] (7-1) Calculate scaling parameters: First, traverse the 3D point cloud matrix of the VIN plane with size (n, 3), and calculate the maximum, minimum, and range of values ​​in the x and y directions respectively. Based on the width and height of the target VIN region, calculate the scaling ratios in the x and y directions. Since the VIN surface may be non-planar or non-parallel, resulting in inconsistent scaling ratios in the two directions, the larger of the two values ​​is selected as a unified scaling factor, and the size of the target image is adjusted accordingly.

[0047] (7-2) Create the target image: Based on the calculated scaling factor, determine the width and height dimensions of the target image (a specific scaling factor must be met to avoid boundary issues caused by rounding), and initialize the target image with a black background. At the same time, calculate the physical dimensions of the normalized target image for reference in subsequent processing.

[0048] (7-3) Projection and Interpolation: Traverse the 3D point cloud data of the VIN plane, mapping the x and y coordinates of each 3D point to the corresponding position in the target image according to the scaling factor. If the coordinates exceed the image boundary, skip them; if the target pixel already has a value, record the number of overlaps for calculating the overlap rate. Using the nearest neighbor interpolation method, assign the RGB data corresponding to the 3D point cloud of the VIN plane to the corresponding pixels in the target image, gradually constructing a 2D texture image. During this process, if the repair function is enabled, a repair mask will be generated simultaneously, marking the positions of the filled pixels.

[0049] (7-4) Post-processing and statistics: Calculate the hole rate, which reflects the proportion of unfilled areas in the 2D texture image. If the inpainting function is enabled, use an inpainting mask to inpaint the 2D texture image and fill in the hole areas. Finally, if visualization assistance is needed, call a function to draw an auxiliary ruler to visually display the scaling information.

[0050] S102. While using a binocular stereo vision camera to photograph the vehicle identification number (VIN), an ultra-high-definition color camera is used to photograph the VIN to obtain a high-resolution color image; based on the color image, an area image containing only the VIN is obtained.

[0051] While the binocular stereo vision camera captures the vehicle identification number (VIN), a high-resolution ultra-high-definition color camera built into the portable terminal is used to capture the VIN to obtain a high-resolution color image. To ensure image quality, the equipment must be kept stable during shooting to avoid image blurring caused by shaking.

[0052] (1) Use a high-resolution ultra-high-definition color camera to photograph the vehicle frame number in order to obtain a high-resolution color image.

[0053] (2) After the captured color image is input into the trained vehicle identification number (VIN) image localization AI model, the model automatically selects the area where the VIN is located through feature extraction and analysis, and crops the image to remove irrelevant background information, so as to obtain an area image containing only the VIN content, providing accurate material for subsequent processing.

[0054] S103. Match the full-scale front view of the vehicle identification number (VIN) with the area image containing only the VIN, and merge them into a high-resolution ultra-high-definition full-scale front view of the VIN.

[0055] Further, step S103 includes: (1) Select feature point matching pairs from the scaled front view of the vehicle frame number and the area image containing only the vehicle frame number to establish the correspondence between the two images.

[0056] An AI model for feature point matching is used to process a scaled-down front view of the vehicle identification number (VIN) and an image containing only the VIN. The model detects key points in the images, such as edges and corners, and extracts feature descriptors for each key point. It then selects feature point pairs with a matching degree higher than a custom threshold to establish a correspondence between the two images.

[0057] Among them, the feature point matching AI model uses the existing network model structure and adopts feature point matching algorithms such as disk / sift / surf / orb.

[0058] (2) Solve the perspective matrix using the RANSAC algorithm for the feature point matching pairs; perform perspective transformation on the region image using the perspective matrix to make it consistent with the perspective of the vehicle frame number in the proportional front view; then replace the color information in the proportional front view of the vehicle frame number with the color information in the transformed region image.

[0059] For the acquired feature point matching pairs, the perspective matrix is ​​calculated using the homography matrix calculation method and the RANSAC algorithm. This perspective matrix describes the planar projection relationship between two images. The VIN region image is then transformed using the perspective matrix to adjust its viewpoint and shape, making it consistent with the viewpoint of the VIN image containing depth information. The color information of the transformed VIN region image then replaces the color information in the depth-information VIN image captured by the stereo camera. Finally, a VIN image with both high-resolution color detail and depth information is obtained, effectively improving image clarity and accuracy.

[0060] (3) Adjust the size of the vehicle identification number (VIN) in the proportional front view that replaces the color information to achieve a one-to-one high-resolution texture restoration of the VIN.

[0061] Based on the high-resolution, scaled-to-scale front view of the vehicle identification number (VIN) output by this trinocular camera system, including parameters such as length and width, the calculated size information (i.e., the physical size of the normalized target image in step S101 (7-2)) is adapted to the specifications of A4 paper. The length and width scaling parameters are adjusted, and the VIN image is printed onto A4 paper at a precise size ratio. This ensures that the VIN image on the A4 paper has clear texture and its size is completely consistent with the original VIN, achieving a high-precision one-to-one reproduction.

[0062] In addition, this embodiment also provides a high-resolution texture restoration system for vehicle identification number (VIN) at a 1:1 ratio, including an image processing module, an image acquisition module, and an image matching and merging module, wherein: (1) Image processing module.

[0063] First, images are simultaneously acquired by the left and right infrared cameras of a binocular vision camera. The depth information of each pixel is calculated using structured light and binocular parallax principles, and then fused with color information to generate an RGBD image. Next, the RGBD image is converted into 3D point cloud data based on the camera intrinsic parameter matrix. Then, the vehicle identification number (VIN) is located using a trained AI model and object detection algorithm. Subsequently, to address invalid depth data and noise interference, valid point cloud data is first filtered, and then the VIN plane normal is calculated using the SVD method. Afterward, the VIN plane normal vector and the camera facing direction vector are normalized, and a rotation matrix is ​​calculated through vector alignment operations. Finally, the rotation matrix is ​​used to rotate and correct the 3D point cloud data, aligning the VIN plane with the camera and providing standard viewpoint data for texture reconstruction.

[0064] (2) Image acquisition module.

[0065] Specifically, it integrates a professional vehicle identification number (VIN) image localization AI model with a feature point matching AI model optimized by advanced algorithms, possessing powerful image recognition, analysis, and processing capabilities. By inputting VIN images captured by an ultra-high-definition color camera, this module can quickly and accurately locate the VIN within the image, perform region cropping, and reduce interference from invalid data.

[0066] (3) Image matching and merging module.

[0067] Specifically, it is used to efficiently detect and match feature points in cropped VIN region images and VIN images with depth information, accurately calculate feature point similarity, and select reliable matching pairs. Finally, based on the matching pairs, a professional algorithm is used to calculate the perspective matrix, perform perspective transformation on the VIN region image, accurately adjust the viewing angle and shape, and accurately fuse high-resolution color information into the VIN scale-up front view realized by the basic binocular stereo vision processing module using the VIN image with depth information. The low-resolution color texture information is replaced with high-resolution color texture information to achieve image optimization and integration, generating a high-resolution VIN scale-up front view image.

[0068] The system also includes a post-processing and printing module. This module integrates a high-precision 3D dimension calculation algorithm with an image output control program. Based on the VIN image data containing depth information, combined with camera parameters and other information, the actual size of the VIN is accurately calculated, and the calculation results are converted into image printing instructions. This drives the printer to output a high-resolution, scaled-down frontal view image of the VIN onto A4 paper, achieving a high-precision, one-to-one reproduction of the VIN's texture and size.

[0069] This embodiment utilizes an ultra-high-definition color camera with high resolution and a large aperture to capture rich color and texture details of the vehicle identification number (VIN); a binocular stereo vision camera employs advanced imaging technology to acquire the VIN's three-dimensional spatial coordinates in real time. Working together, these two cameras can capture high-resolution color images with depth information.

[0070] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0071] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0072] This embodiment also provides a terminal device, which can be a computer, such as... Figure 2 As shown, it is connected via system bus 201 to processor 202, memory, input device 203, display 204 and network interface 205. The processor provides computing and control capabilities. The memory includes non-volatile storage medium 206 and internal memory 207. The non-volatile storage medium 206 stores the operating system, computer programs and database. The internal memory 207 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When processor 202 executes the computer program stored in memory, it implements the above-mentioned one-to-one high-resolution texture restoration method for vehicle identification number.

[0073] This embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for high-resolution texture restoration of vehicle identification number (VIN) at a ratio of 1:1.

[0074] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0075] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for high-resolution texture reconstruction of vehicle identification number (VIN) at a 1:1 ratio, characterized in that, An application is made to a trinocular stereo vision camera system, which includes a binocular stereo vision camera and an ultra-high-definition color camera. The binocular stereo vision camera includes two infrared cameras, an infrared structured light projector, and a high-definition color camera. The two infrared cameras and the structured light projector work together to acquire depth information of the vehicle identification number (VIN), and the high-definition color camera provides color information of the VIN. The method includes: The vehicle identification number (VIN) is captured from multiple angles using a binocular stereo camera; the captured images are processed to generate a scaled-down orthographic view of the VIN. While using a binocular stereo vision camera to photograph the vehicle identification number (VIN), an ultra-high-definition color camera is also used to photograph the VIN to obtain a high-resolution color image; based on the color image, an area image containing only the VIN is obtained; wherein, the color image overlaps with the image captured by the binocular stereo vision camera by at least half. Match the VIN scale front view with the area image containing only the VIN and merge them into a high-resolution ultra-high-definition VIN scale front view.

2. The method for high-resolution texture reconstruction of vehicle identification number (VIN) at a 1:1 ratio according to claim 1, characterized in that, The process of matching the scaled-to-scale front view of the vehicle identification number (VIN) with an area image containing only the VIN and merging them into a high-resolution, ultra-high-definition scaled-to-scale front view of the VIN includes: Feature point matching pairs are selected from the scaled front view of the vehicle identification number and the area image containing only the vehicle identification number to establish the correspondence between the two images; The perspective matrix is ​​solved using the RANSAC algorithm for feature point matching pairs; the perspective transformation is performed on the region image containing only the vehicle identification number (VIN) using the perspective matrix to make it consistent with the perspective of the VIN's scaled front view; and the color information in the scaled front view of the VIN is then replaced with the color information in the transformed region image. The dimensions of the VIN (Vehicle Identification Number) in a scaled-down front view after the color information has been replaced are adjusted to achieve a one-to-one high-resolution texture reproduction of the VIN.

3. The method for high-resolution texture reconstruction of vehicle identification number (VIN) at a 1:1 ratio according to any one of claims 1 and 2, characterized in that, The process of processing the captured image to generate a scaled-down front view of the vehicle identification number includes: Generate a color depth map based on the captured images; Based on the pre-calibrated camera intrinsic parameter matrix, each pixel in the color depth map is converted into xyz coordinates in the camera coordinate system to construct 3D point cloud data; Based on the converted color depth map, locate the vehicle identification number area; Based on the 3D point cloud data of the vehicle identification number (VIN) region, the SVD algorithm is used to suppress noise interference and extract the VIN plane normal vector. Calculate the rotation matrix facing the camera based on the normal vector of the vehicle identification number plane; Based on the rotation matrix, the vehicle identification number plane is rotated and corrected to face the camera. Based on the rotated and corrected VIN plane, a two-dimensional texture image is constructed using ray projection to generate a scaled-down front view of the VIN.

4. The method for high-resolution texture restoration of vehicle identification number (VIN) at a 1:1 ratio according to claim 3, characterized in that, The step of constructing a two-dimensional texture image based on the rotated and corrected vehicle identification number (VIN) plane using light projection includes: Calculate the width and height of the VIN area based on the VIN area where the rotated and corrected VIN plane is located; Calculate the scaling ratio based on the calculated width and height and the width and height of the target VIN area; Based on the scaling ratio, determine the width and height of the target image, and initialize the target image with a black background; The process iterates through the 3D point cloud data of the VIN region where the VIN plane is located after rotation correction, mapping the x and y coordinates of each 3D point to the corresponding position in the target image according to the scaling factor; if the x and y coordinates exceed the boundary of the target image, they are skipped; if the target pixel already has a value, the number of overlaps is recorded for calculating the overlap rate; using the nearest neighbor interpolation method, the RGB data corresponding to the 3D point cloud of the VIN region are assigned to the corresponding pixels in the target image, gradually constructing a two-dimensional texture image to generate a scaled front view of the VIN.

5. The method for high-resolution texture reconstruction of vehicle identification number (VIN) at a 1:1 ratio according to claim 4, characterized in that, Repair masks are used to repair the scaled front view of the generated vehicle frame number to fill in the empty areas.

6. The method for high-resolution texture reconstruction of vehicle identification number (VIN) at a 1:1 ratio according to claim 3, characterized in that, The step of using the SVD algorithm to suppress noise interference and extract the normal vector of the VIN plane based on the 3D point cloud data of the VIN region includes: Based on the set depth value filtering criteria, the 3D point cloud data of the vehicle identification number area is filtered to select valid 3D point cloud data. The SVD method is used to decompose the 3D point cloud data of the filtered vehicle identification number area into a product of matrices U, Σ, and Vt. Transpose matrix U to obtain matrix Ut; find the index of the element with the largest absolute value in each row of matrix Ut and arrange them into a column, denoted as max_abs_u_cols; create an arithmetic sequence shift from 0 to the number of rows in matrix Ut minus 1; multiply shift by the number of columns in matrix Ut and add it to each element of max_abs_u_cols to obtain indices; extract elements from matrix Ut based on indices and obtain the signs of these elements, denoted as signs; Multiply each element in signs by the corresponding element in the corresponding column of matrix Vt to obtain the adjusted matrix Vt; extract the last row of data from the adjusted matrix Vt as the normal vector of the chassis number plane.

7. The method for high-resolution texture reconstruction of vehicle identification number (VIN) at a 1:1 ratio according to claim 3, characterized in that, The step of generating a color depth map based on the captured image includes: Images of the vehicle identification number (VIN) are simultaneously acquired by two infrared cameras on the left and right sides. The depth information of each pixel in the image is calculated using structured light or binocular parallax principle. While acquiring VIN images simultaneously by the two cameras, images are also acquired by a high-definition color camera. The color information and depth information of each pixel in the image are fused to generate a color depth map.

8. A vehicle identification number (VIN) 1:1 high-resolution texture reproduction system, characterized in that, This system is applied to a trinocular stereo vision camera system, which includes a binocular stereo vision camera and an ultra-high-definition color camera. The binocular stereo vision camera comprises two infrared cameras, an infrared structured light projector, and a high-definition color camera. The two infrared cameras and the structured light projector work together to acquire depth information of the vehicle identification number (VIN). The high-definition color camera provides color information for the VIN. The VIN 1:1 high-resolution texture reconstruction system includes: The image processing module is used to capture the vehicle identification number (VIN) from multiple angles using a binocular stereo vision camera; and to process the captured images to generate a scaled-down orthographic view of the VIN. The image acquisition module is used to simultaneously capture the vehicle identification number (VIN) using a binocular stereo vision camera and an ultra-high-definition color camera to obtain a high-resolution color image; based on the color image, a region image containing only the VIN is obtained; wherein the color image and the image captured by the binocular stereo vision camera overlap by at least half. The image matching and merging module is used to match the full-scale front view of the vehicle identification number (VIN) with an area image containing only the VIN, and merge them into a high-resolution, ultra-high-definition full-scale front view of the VIN.

9. A terminal device, comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the vehicle identification number (VIN) one-to-one high-resolution texture restoration method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the vehicle identification number (VIN) one-to-one high-resolution texture restoration method according to any one of claims 1 to 7.

Images