A depth-guided, disparity-constrained binocular structured light stereo matching acceleration method

By employing a depth-guided parallax constraint method, this method utilizes monocular structured light to generate parallax prior constraints and combines them with dual matching verification. This solves the problems of low efficiency and mismatch in binocular structured light measurement systems at high resolution, achieving efficient 3D reconstruction and high-quality point cloud generation.

CN123118458APending Publication Date: 2026-11-17NORTHWESTERN POLYTECHNICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611091768.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing binocular structured light measurement systems have low stereo matching efficiency at high resolutions, and traditional methods produce incorrect matching in non-common field of view areas, resulting in false noise at the edges of the reconstructed point cloud, which affects measurement accuracy and efficiency.

Method used

A depth-guided disparity constraint method is adopted, which generates the disparity prior constraint search range through monocular structured light reconstruction, and combines dual matching constraint verification to eliminate mismatches, thereby accelerating stereo matching in high-resolution scenes.

Benefits of technology

It significantly improves stereo matching efficiency, suppresses point cloud edge mismatch, enhances 3D reconstruction quality, and meets the needs of industrial online rapid inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118458A_ABST
    Figure CN123118458A_ABST
Patent Text Reader

Abstract

This invention discloses a depth-guided, disparity-constrained binocular structured light stereo matching acceleration method. It utilizes left and right cameras to construct monocular structured light systems with a projector, respectively, to reconstruct left and right depth maps. Stereo correction is performed on the left and right absolute phase maps, and the 3D points reconstructed by the left and right monocular structured light systems are transformed to their corresponding corrected camera coordinate systems and reprojected to generate left and right corrected depth maps. Based on the mapping relationship between depth values ​​and disparity, initial left and right disparity maps are generated. The forward local search interval is limited by the left initial disparity value, and candidate points are screened using phase matching validity verification. The reverse local search interval is then limited by the right initial disparity value, and valid corresponding points are obtained through reverse phase matching and bidirectional matching consistency verification. Subpixel disparity is calculated, and binocular 3D reconstruction is completed. This invention can significantly reduce the time required for 3D reconstruction of high-resolution images while effectively suppressing mismatches, balancing reconstruction efficiency and point cloud quality.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of structured light 3D measurement technology, specifically relating to a depth-guided disparity-constrained binocular structured light stereo matching acceleration method. Background Technology

[0002] Structured light 3D measurement technology, with its advantages of being non-contact, highly accurate, and having a high sampling rate, has become one of the core technologies in fields such as industrial precision inspection and reverse engineering. Among them, the binocular structured light measurement system combines active structured light coding with the principle of binocular vision triangulation, and has higher measurement accuracy and robustness than the monocular structured light system, making it the mainstream technical solution for high-precision 3D measurement.

[0003] In the binocular structured light measurement process, stereo matching is the core step for determining corresponding feature points in the left and right cameras and achieving 3D reconstruction; it is also the most time-consuming part of the overall algorithm. Existing binocular stereo matching methods based on phase features typically use binocular stereo correction to align the epipolar lines to the image row direction, and then search pixel by pixel in the corresponding row of the right image for feature points with the same phase as the point to be matched in the left image—that is, a row-by-row full-range search. This method can meet the requirements for rapid detection with low-resolution cameras, but with the increasing demands for resolution and accuracy in industrial inspection, 4K and higher resolution industrial cameras are widely used. The time complexity of the row-by-row full-range search is O(n³), and matching a single frame image requires billions of calculations, which severely restricts measurement efficiency and makes it difficult to meet the needs of rapid industrial inspection.

[0004] Due to the difference in field of view between binocular cameras and projectors, the surface of the object being measured contains non-common coded regions that are visible only to a single camera or shared by both a single camera and the projector. Traditional line-by-line matching algorithms do not effectively constrain the matching range, and the phase features of these non-common regions generate a large number of incorrect matches. This results in a ring of false noise at the edge of the reconstructed point cloud, which not only increases the cost of subsequent point cloud denoising but also introduces measurement errors, affecting the final measurement accuracy.

[0005] To improve stereo matching efficiency, existing technologies mostly employ hardware acceleration solutions, such as multi-threaded parallel computing and GPU acceleration, but these significantly increase system hardware costs and deployment complexity. Some algorithms reduce the search range through image downsampling, but this directly sacrifices matching accuracy, failing to meet the requirements of precision measurement. On the other hand, monocular structured light systems offer fast reconstruction speeds, but their measurement accuracy is relatively low, making it difficult to meet high-precision measurement needs on its own. Existing technologies have failed to effectively combine the speed advantage of monocular structured light with the accuracy advantage of binocular structured light. Therefore, there is an urgent need to propose a binocular structured light measurement scheme that can significantly improve stereo matching efficiency while suppressing edge mismatches, while ensuring binocular measurement accuracy. Summary of the Invention

[0006] This invention addresses the problems of existing methods by providing a depth-guided, disparity-constrained binocular structured light stereo matching acceleration method. This method leverages the advantages of fast reconstruction speed with monocular structured light and high measurement accuracy with binocular structured light. It converts the monocular depth map into a disparity prior constraint search range and eliminates mismatches through dual matching constraint verification. This achieves efficient acceleration of stereo matching in high-resolution scenes, improves point cloud reconstruction quality, and suppresses edge noise interference caused by mismatches in non-common fields of view.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: (1) Control the projector to project a structured light coded image onto the object under test, and acquire corresponding coded images through the left and right cameras respectively. Decode the acquired coded images to obtain the left absolute phase image from the left camera perspective and the right absolute phase image from the right camera perspective. (2) Based on the binocular structured light system, a left monocular structured light subsystem consisting of a left camera and a projector and a right monocular structured light subsystem consisting of a right camera and a projector are constructed respectively. Monocular structured light 3D reconstruction is performed respectively to obtain the left depth map in the left camera coordinate system and the right depth map in the right camera coordinate system. (3) According to the stereo correction parameters, the left and right absolute phase maps are stereo corrected. The three-dimensional points obtained by the three-dimensional reconstruction of the left and right monocular structured light are transformed to the left and right camera coordinate systems after stereo correction and projected onto the left and right camera correction image planes to obtain the left correction depth map and the right correction depth map respectively. According to the mapping relationship between the correction depth value and the stereo disparity, the initial disparity value of the corresponding pixels in the left and right images is calculated to generate the left and right initial disparity maps. (4) For the pixel to be matched in the corrected left absolute phase map, the forward local disparity search interval is determined according to its left initial disparity value. Phase matching and phase matching validity verification are performed in the forward local disparity search interval to obtain forward matching candidate pixels. The reverse local disparity search interval is determined according to the right initial disparity value of the forward matching candidate pixels. Reverse phase matching and bidirectional consistency verification are performed. The forward matching candidate pixels that pass the bidirectional consistency verification are determined as valid corresponding points. (5) Calculate subpixel disparity based on the phase information of effective corresponding points, and perform three-dimensional reconstruction according to the subpixel disparity and binocular reprojection relationship to generate point cloud of the surface of the measured object.

[0008] Furthermore, in step (2) above, the specific steps for establishing the monocular structured light depth map are as follows: 1) Treat the projector as an equivalent of a reverse camera; 2) Determine the corresponding projector image coordinates based on the absolute phase value of the measured pixel; 3) Based on the camera image coordinates, projector image coordinates, camera parameters, projector parameters, and the relative pose parameters between the camera and the projector, solve the camera projection equation and the projector projection equation simultaneously to solve for the three-dimensional coordinates and depth values ​​of the corresponding pixels. 4) Store the depth values ​​according to the image pixel positions of the corresponding cameras to form a left depth map and a right depth map.

[0009] Furthermore, in step (3) above, the specific steps for generating the left initial disparity map and the right initial disparity map are as follows: 1) Transform the left monocular 3D points obtained from the left monocular structured light 3D reconstruction to the left camera coordinate system after stereo correction, and obtain the left corrected depth value corresponding to the monocular 3D points. Based on the existing binocular camera calibration parameters, the pixels in the corrected left camera image are calculated using the following formula. The corresponding initial left disparity value: in, f B is the equivalent focal length of the camera in the horizontal direction after stereo correction, and B is the baseline length. 2) Transform the right monocular 3D points obtained from the right monocular structured light 3D reconstruction to the right camera coordinate system after stereo correction, and obtain the right corrected depth value corresponding to the right monocular 3D points. Based on the existing binocular camera calibration parameters, the pixels in the corrected right camera image are calculated using the following formula. The corresponding initial right disparity value: 3) Store the left and right initial disparity values ​​according to the positions of the corresponding pixels in the corrected left and right camera images to generate the left initial disparity map and the right initial disparity map.

[0010] Furthermore, in step (4) above, the forward local disparity search interval and the reverse local disparity search interval are respectively: in, and These are the search radii for the forward local disparity search interval and the reverse local disparity search interval, respectively.

[0011] The search radius is a preset fixed pixel width, or it can be adaptively determined based on at least one of the following: monocular depth uncertainty of the corresponding pixel, system calibration error, phase position confidence, phase decoding error, and depth gradient.

[0012] Furthermore, in step (4) above, the phase matching validity verification specifically involves: 1) Within the positive local disparity search interval, calculate the absolute phase difference between the absolute phase value of each search pixel and the absolute phase value of the pixel to be matched, and determine the search pixel with the smallest absolute phase difference as the positive matching candidate pixel. 2) When the minimum absolute phase difference corresponding to the positive matching candidate pixel is less than the preset phase difference threshold, the positive matching candidate pixel is determined to be a positive matching pixel; when the minimum absolute phase difference is not less than the preset phase difference threshold, the pixel to be matched is determined to be a non-positive matching pixel.

[0013] Furthermore, in step (4) above, the bidirectional consistency verification specifically involves: 1) Using the positive matching candidate pixel that has passed the phase matching validity check as the negative matching pixel, determine the negative local disparity search interval centered on the right initial disparity value according to the right initial disparity value of the negative matching pixel in the right initial disparity map. 2) On the left camera corrected image epipolar line corresponding to the reverse local parallax search interval, calculate the absolute phase difference between each candidate pixel and the reverse matching pixel, determine the candidate pixel with the smallest absolute phase difference as the reverse matching candidate pixel, and determine whether the smallest absolute phase difference is less than the preset reverse phase difference threshold. 3) When the minimum absolute phase difference is less than the preset reverse phase difference threshold, the reverse matching candidate pixel is determined as the reverse matching pixel, and the lateral coordinate deviation between the reverse matching pixel and the original pixel to be matched in the left camera correction image in the epipolar direction is calculated. 4) When the lateral coordinate deviation is not greater than the preset pixel deviation threshold, the positive matching candidate pixel is determined to have passed the bidirectional consistency check; when the minimum absolute phase difference is not less than the preset reverse phase difference threshold, or the lateral coordinate deviation is greater than the preset pixel deviation threshold, the positive matching candidate pixel is determined to have failed the bidirectional consistency check.

[0014] Furthermore, in step (5) above, the subpixel disparity is calculated as follows: based on the phase values ​​of the candidate pixel and its neighboring pixels, the matching position with subpixel precision is obtained by phase interpolation fitting, and then the subpixel disparity value is calculated.

[0015] The beneficial effects of this invention are: This invention provides a depth-guided parallax-constrained binocular structured light stereo matching acceleration method, which can significantly reduce the time required for 3D reconstruction of high-resolution images, effectively suppress false noise at the edge of point clouds generated by traditional matching algorithms, and can adapt to the application requirements of industrial online rapid detection, thus simultaneously improving the overall performance of binocular structured light 3D reconstruction in terms of both efficiency and quality. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention.

[0017] Figure 2 This is a schematic diagram of a binocular structured light measurement system and its left and right monocular structured light subsystems.

[0018] Figure 3 This is a comparison of the reconstructed point clouds from two matching algorithms. Detailed Implementation

[0019] The principles and features of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0020] As attached Figure 1 As shown, the depth-guided parallax-constrained binocular structured light stereo matching acceleration method of the present invention uses a binocular structured light measurement system including a left camera, a right camera, and a projector.

[0021] As attached Figure 2 As shown, in this embodiment, both the left and right cameras are industrial cameras with a resolution of 4504×4504, symmetrically arranged in front of the object being measured. The projector is a DLP digital light processing projector with a resolution of 1920×1080, located in the middle of the binocular cameras. The left and right cameras are triggered by the projector to synchronously acquire projected images. Before measurement, the calibration between the monocular camera, the projector, and the binocular camera is completed in advance. The specific steps for accelerating the reconstruction of three-dimensional point clouds using the method of this invention are as follows.

[0022] (1) In this embodiment, the projector is first controlled to project a structured light coded image onto the object being measured. The structured light coded image includes a coding pattern of phase shift method combined with Gray code and complementary Gray code set along the horizontal and vertical directions of the projector, respectively. When the projector projects each frame of coded image, the left camera and the right camera synchronously acquire the corresponding image, and obtain the left camera coded image sequence and the right camera coded image sequence, respectively.

[0023] Using pre-calibrated camera distortion parameters, distortion correction is performed on the left and right coded image sequences respectively to eliminate the effects of radial and tangential lens distortion. The corrected images are then decoded: the wrap-around phase is calculated using multiple phase-shifted images. Then, combined with the complementary Gray code encoding results, the package phase is... Perform periodic unwrapping to obtain the left absolute phase image from the left camera's perspective. Right absolute phase diagram from the perspective of the right camera .

[0024] (2) In this embodiment, the left camera and the projector constitute a left monocular structured photonic system, and the right camera and the projector constitute a right monocular structured photonic system.

[0025] In a left-side monocular structured photonics system, for the effective coded pixels in the left camera image... Its image coordinates can be represented as According to the absolute phase map of the pixel on the left The absolute phase value can be used to determine the corresponding projector image coordinates. By treating the projector as an equivalent inverted camera, and based on the projection equations of the left camera and the projector, the following projection constraint relationship can be established: Where P is a three-dimensional point on the surface of the object being measured. This represents the projection model of the left camera. This represents the projection model of the projector. The coordinates of the projector image are given. Based on the intrinsic parameters of the left camera, the intrinsic parameters of the projector, and the relative pose parameters between the left camera and the projector, the coordinates and depth value of the 3D point P can be solved simultaneously.

[0026] Similarly, in a right-side monocular structured photonic system, based on the right camera image coordinates and the right absolute phase map... Given the corresponding projector image coordinates, right camera intrinsic parameters, projector intrinsic parameters, and relative pose parameters between the right camera and the projector, calculate the depth value of the corresponding pixel from the right camera's perspective.

[0027] The depth values ​​obtained from the left camera's viewpoint are stored according to the pixel positions of the left camera image to obtain the left depth map. The depth values ​​obtained from the right camera's viewpoint are stored according to the pixel positions of the right camera image to obtain the right depth map. .

[0028] (3) Based on the pre-calibrated binocular stereo correction parameters, adjust the left absolute phase map. Right absolute phase diagram Stereo correction is performed by transforming the 3D points obtained from the left and right monocular structured light 3D reconstructions to the left and right camera coordinate systems after stereo correction and projecting them onto the left camera correction image plane to obtain the left correction depth map and the right correction depth map. The left correction depth map is combined with the mapping relationship between depth values ​​and binocular disparity to calculate the initial disparity values ​​of corresponding pixels in the left and right images, and generate the left and right initial disparity maps. The specific steps for generating the left initial disparity map and the right initial disparity map are as follows.

[0029] 1) Transform the left monocular 3D points obtained from the left monocular structured light 3D reconstruction to the left camera coordinate system after stereo correction, and obtain the left corrected depth value corresponding to the monocular 3D points. Based on the existing binocular camera calibration parameters, the pixels in the corrected left camera image are calculated using the following formula. The corresponding initial left disparity value: in, f B is the equivalent focal length of the camera in the horizontal direction after stereo correction, and B is the baseline length. This represents the left-corrected depth value corresponding to a monocular 3D point.

[0030] 2) Transform the right monocular 3D points obtained from the right monocular structured light 3D reconstruction to the right camera coordinate system after stereo correction, and obtain the right corrected depth value corresponding to the right monocular 3D points. Based on the existing binocular camera calibration parameters, the pixels in the corrected right camera image are calculated using the following formula. The corresponding initial right disparity value: in, This represents the right-corrected depth value corresponding to the right monocular 3D point.

[0031] 3) Store the left and right initial disparity values ​​according to the positions of the corresponding pixels in the corrected left and right camera images to generate the left initial disparity map and the right initial disparity map.

[0032] (4) For the pixel to be matched in the corrected left absolute phase map, the forward local disparity search interval is determined based on its left initial disparity value in the left initial disparity map. In this embodiment, the forward local disparity search radius is... Set to a fixed 3 pixels, therefore, without being truncated by image boundaries, the forward local disparity search interval contains 7 integer disparity candidates centered on the left initial disparity value. The forward local disparity search interval is: in, These are the pixels to be matched in the corrected left absolute phase map.

[0033] On the epipolar line of the right camera corrected image corresponding to the positive local disparity search interval, the absolute phase difference between each candidate pixel and the pixel to be matched in the left camera corrected image is calculated, and the candidate pixel with the smallest absolute phase difference is determined as the positive matching candidate pixel.

[0034] It is determined whether the minimum absolute phase difference corresponding to the forward matching candidate pixel is less than a preset phase difference threshold. In this embodiment, the preset phase difference threshold is set to 0.25 rad. When the minimum absolute phase difference is less than 0.25 rad, the forward matching candidate pixel is determined to have passed the phase matching validity check; otherwise, it is determined that there is no valid forward matching candidate pixel for the current pixel to be matched, and subsequent reverse matching and 3D reconstruction are no longer performed.

[0035] For the positive matching candidate pixel that passes the phase matching validity check, the positive matching candidate pixel is used as the negative matching pixel, and the negative local disparity search interval is determined according to its right initial disparity value in the right initial disparity map. In this embodiment, the negative local disparity search radius is... Also set to 3 pixels. The inverse local disparity search interval is: in, The reverse pixel to be matched.

[0036] On the left camera corrected image epipolar line corresponding to the reverse local parallax search interval, the absolute phase difference between each candidate pixel and the reverse matching pixel is calculated respectively. The candidate pixel with the smallest absolute phase difference is determined as the reverse matching candidate pixel, and it is determined whether the reverse matching candidate pixel is valid.

[0037] In this embodiment, the preset phase difference threshold used for reverse phase matching is 0.25 rad. When the minimum absolute phase difference corresponding to a reverse matching candidate pixel is less than 0.25 rad, it is determined to be a reverse matching pixel; otherwise, the corresponding forward matching candidate pixel is determined to have failed the bidirectional consistency check and is removed.

[0038] For a valid reverse-matching pixel, its lateral coordinate deviation in the epipolar direction from the original pixel to be matched in the corrected image of the left camera is calculated. In this embodiment, the preset pixel deviation threshold is set to 3 pixels. When the lateral coordinate deviation is no greater than 3 pixels, the forward-matching candidate pixel is determined to have passed the bidirectional consistency check, and the original pixel to be matched and the forward-matching candidate pixel are identified as valid corresponding points; otherwise, the forward-matching candidate pixel is determined to be a mismatched pixel and is discarded.

[0039] By performing the above-mentioned forward phase matching validity check and bidirectional consistency check, on the one hand, the search range of forward and reverse phase matching can be narrowed by utilizing the initial disparity constraint, thereby reducing the computational load of phase matching; on the other hand, erroneous matching results that cannot return to the original matching position after forward matching can be eliminated, thereby reducing invalid matching.

[0040] (5) For valid corresponding points that pass the phase matching validity check and bidirectional consistency check, phase interpolation fitting is performed based on the absolute phase values ​​of the valid corresponding points and their neighboring pixels to obtain the sub-pixel precision matching position, and the sub-pixel disparity is calculated based on the difference between the horizontal coordinate of the pixel to be matched in the left camera correction image and the horizontal coordinate of the sub-pixel matching position in the right camera correction image.

[0041] Based on the reprojection relationship obtained from the subpixel parallax and binocular stereo correction, the three-dimensional coordinates of the points on the surface of the object under test are calculated, and a point cloud of the surface of the object under test is generated from each effective three-dimensional point.

[0042] To verify the efficiency advantage of the method of this invention, a three-dimensional point cloud reconstruction of a freeform surface sample was performed using a line-by-line stereo matching method based on binocular stereo correction constraints and the method of this invention. The running time of the stereo matching step was statistically analyzed. Table 1 shows the running time of the two matching algorithms.

[0043] The 3D reconstruction speed of this invention is 13.03 times that of the traditional binocular stereo correction constraint matching algorithm, and the stereo matching time is reduced by 92.33%. It can be seen that the method of this invention can significantly improve the efficiency of binocular structured light 3D reconstruction.

[0044] Table 1 Matching Algorithm Runtime (ms) Line-by-line stereo matching algorithm based on binocular stereo correction constraints 14241.60 Depth-guided disparity-constrained binocular stereo matching acceleration algorithm 1092.93 Appendix Figure 3 The image shows a comparison of the point clouds of 3D reconstructed freeform surface samples using the traditional line-by-line stereo matching method and the method of this invention. Figure 3 (a) shows the reconstruction result of the traditional line-by-line stereo matching method. Figure 3 (b) shows the reconstruction result of the method of the present invention. It can be seen that the method of the present invention can effectively suppress mismatched noise points in the point cloud and significantly improve the overall quality of the reconstructed point cloud.

[0045] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A depth-guided, parallax-constrained binocular structured light stereo matching acceleration method, applied to a binocular structured light measurement system consisting of a left camera, a right camera, and a projector, characterized in that... The method includes the following steps: (1) Control the projector to project a structured light coded image onto the object under test, and acquire corresponding coded images through the left and right cameras respectively. Decode the acquired coded images to obtain the left absolute phase image from the left camera perspective and the right absolute phase image from the right camera perspective. (2) Based on the binocular structured light system, construct a left monocular structured light subsystem consisting of a left camera and a projector and a right monocular structured light system consisting of a right camera and a projector, respectively, and perform monocular structured light 3D reconstruction to obtain the left depth map in the left camera coordinate system and the right depth map in the right camera coordinate system. (3) Based on the stereo correction parameters, stereo correction is performed on the left and right absolute phase maps. The three-dimensional points obtained by the three-dimensional reconstruction of the left and right monocular structured light are transformed to the left and right camera coordinate systems after stereo correction and projected onto the left and right camera correction image planes to obtain the left correction depth map and the right correction depth map respectively. Based on the mapping relationship between the correction depth value and the stereo disparity, the initial disparity values ​​of the corresponding pixels in the left and right images are calculated to generate the left and right initial disparity maps. (4) For the pixel to be matched in the corrected left absolute phase map, the forward local disparity search interval is determined according to its left initial disparity value. Phase matching and phase matching validity verification are performed within the forward local disparity search interval to obtain forward matching candidate pixels. The reverse local disparity search interval is determined according to the right initial disparity value of the forward matching candidate pixels. Reverse phase matching and bidirectional consistency verification are performed. The forward matching candidate pixels that pass the bidirectional consistency verification are determined as valid corresponding points. (5) Calculate subpixel disparity based on the phase information of effective corresponding points, and perform three-dimensional reconstruction by combining binocular reprojection relationship to generate point cloud of the surface of the measured object.

2. The depth-guided disparity-constrained binocular structured light stereo matching acceleration method as described in claim 1, characterized in that, In step (2), the specific steps for establishing the monocular structured light depth map are as follows: 1) Treat the projector as an equivalent of a reverse camera; 2) Determine the corresponding projector image coordinates based on the absolute phase value of the measured pixel; 3) Based on the camera image coordinates, projector image coordinates, camera parameters, projector parameters, and the relative pose parameters between the camera and the projector, solve the camera projection equation and the projector projection equation simultaneously to solve for the three-dimensional coordinates and depth values ​​of the corresponding pixels. 4) Store the depth values ​​according to the corresponding image pixel positions of the camera to form a left depth map and a right depth map.

3. The depth-guided disparity-constrained binocular structured light stereo matching acceleration method as described in claim 1, characterized in that, In step (3), the specific steps for generating the left initial disparity map and the right initial disparity map are as follows: 1) Transform the left monocular 3D points obtained from the left monocular structured light 3D reconstruction to the left camera coordinate system after stereo correction, and obtain the left corrected depth value corresponding to the monocular 3D points. Based on the existing binocular camera calibration parameters, the pixels in the corrected left camera image are calculated using the following formula. Corresponding left initial disparity value: Where f is the equivalent focal length of the camera in the horizontal direction after stereo correction, and B is the baseline length. 2) Transform the right monocular 3D points obtained from the right monocular structured light 3D reconstruction to the right camera coordinate system after stereo correction, and obtain the right corrected depth value corresponding to the right monocular 3D points. Based on the existing binocular camera calibration parameters, the pixels in the corrected right camera image are calculated using the following formula. The corresponding right initial disparity value: 3) Store the left and right initial disparity values ​​according to the positions of the corresponding pixels in the corrected left and right camera images to generate the left initial disparity map and the right initial disparity map.

4. The depth-guided disparity-constrained binocular structured light stereo matching acceleration method as described in claim 1, characterized in that, In step (4), the forward local disparity search interval and the reverse local disparity search interval are respectively: in, and These are the search radii for the forward local disparity search interval and the reverse local disparity search interval, respectively.

5. The depth-guided disparity-constrained binocular structured light stereo matching acceleration method as described in claim 4, characterized in that, The search radius is a preset fixed pixel width, or it can be adaptively determined based on at least one of the following: monocular depth uncertainty of the corresponding pixel, system calibration error, phase position confidence, phase decoding error, and depth gradient.

6. The depth-guided disparity-constrained binocular structured light stereo matching acceleration method as described in claim 1, characterized in that, In step (4), the phase matching validity verification specifically includes: 1) Calculate the absolute phase difference between the absolute phase value of each search pixel in the positive local disparity search interval and the absolute phase value of the pixel to be matched, and determine the search pixel with the smallest absolute phase difference as the positive matching candidate pixel; 2) When the minimum absolute phase difference corresponding to the positive matching candidate pixel is less than the preset phase difference threshold, the positive matching candidate pixel is determined to be a positive matching pixel; when the minimum absolute phase difference is not less than the preset phase difference threshold, the pixel to be matched is determined to be a non-positive matching pixel.

7. The depth-guided disparity-constrained binocular structured light stereo matching acceleration method as described in claim 1, characterized in that, In step (4), the bidirectional consistency verification specifically includes: 1) Using the positive matching candidate pixel that has passed the phase matching validity check as the negative matching pixel, determine the negative local disparity search interval centered on the right initial disparity value according to the right initial disparity value of the negative matching pixel in the right initial disparity map. 2) On the left camera corrected image epipolar line corresponding to the reverse local parallax search interval, calculate the absolute phase difference between each candidate pixel and the reverse matching pixel, determine the candidate pixel with the smallest absolute phase difference as the reverse matching candidate pixel, and determine whether the smallest absolute phase difference is less than the preset reverse phase difference threshold. 3) When the minimum absolute phase difference is less than the preset reverse phase difference threshold, the reverse matching candidate pixel is determined as the reverse matching pixel, and the lateral coordinate deviation between the reverse matching pixel and the original pixel to be matched in the left camera correction image in the epipolar direction is calculated. 4) When the lateral coordinate deviation is not greater than the preset pixel deviation threshold, the positive matching candidate pixel is determined to have passed the bidirectional consistency check; when the minimum absolute phase difference is not less than the preset reverse phase difference threshold, or the lateral coordinate deviation is greater than the preset pixel deviation threshold, the positive matching candidate pixel is determined to have failed the bidirectional consistency check.

8. The depth-guided disparity-constrained binocular structured light stereo matching acceleration method as described in claim 1, characterized in that, In step (5), the subpixel disparity is calculated as follows: based on the phase values ​​of the candidate pixel and its neighboring pixels, the matching position with subpixel precision is obtained by phase interpolation fitting, and then the subpixel disparity value is calculated.