Research On Distance Measurement And SLAM Method Based On Binocular Vision

As one of the popular research fields of machine vision,binocular vision is widely used in navigation and obstacle avoidance,3D reconstruction,virtual reality,and other fields.Simultaneous Localization and Mapping(SLAM)has become an important technology in the field of binocular vision and has made great progress,but the complexity and variability of the scene as well as the presence of occlusions and weak textures pose great challenges to the accuracy of binocular ranging and SLAM map building.To this end,this paper addresses the above issues by investigating stereo matching for binocular ranging and the creation of key frames in SLAM as follows:(1)A stereo matching algorithm for improved binocular ranging is proposed(the algorithm in this paper).In the calculation process of Sum of Absolute Difference(SAD)algorithm,the Image Difference Method(IDM)is used to effectively eliminate redundant calculations and reduce the complexity of the algorithm;in the Census algorithm,a bidirectional search method is used to reduce the memory occupation of data and shorten the algorithm operation time.The two improved algorithms are integrated to enhance the antiinterference capability of the system,and the edge detection operator is used to constrain the image edges,the adaptive penalty coefficient is designed to punish the non-continuous areas of the image edges,and the edge parallax is smoothed by guided filtering.Under the Middlebury dataset,the algorithm in this paper reduces the average mis-matching rate by4.08% compared with the Semi-Global Matching(SGM)algorithm,shortens the average operation time by 3.5s compared with the SGM algorithm,has a better parallax map effect compared with the BM algorithm,has smoother edge regions,and accurately restores the depth information of objects in the scene.The UAV ranging accuracy is maintained within96.5%,which provides preparation for subsequent key frame screening based on feature depth in binocular SLAM.(2)An improved binocular SLAM algorithm based on feature depth screening is proposed(OUR-ORB-SLAM3 algorithm).In the feature extraction process,the adaptive corner point detection algorithm is used to obtain more feature points by reducing the threshold value;the optical flow method is used to track the image features,and the Prosac algorithm is used to calculate the updated single response matrix to form the compensation and mis-match rejection of key points;the image features are associated with the depth information as the basis for the selection of key frames,and the current data are associated using local map building threads,and for the duplicate The map points are fused to reduce the computation and optimize the execution efficiency of the algorithm.Under the KITTI dataset,the OUR-ORB-SLAM3 algorithm improves the root mean square error by nearly55% compared with the ORB-SLAM3 algorithm in the optimal case.With 11 binocular sequences in the EuRoc dataset,the OUR-ORB-SLAM3 algorithm was able to obtain smaller mean square error under 6 of them,and was able to extract more stable feature information with higher system robustness.