Research On Road SLAM Technology Based On Binocular Vision

In recent years,as one of the research field of artificial intelligence,the autonomous driving technology has made amazing progress.Simultaneous Localization and Mapping(SLAM)is one of the major challenges faced by autonomous vehicles.With the development of computer vision technology,visual SLAM has become a new hot research field,which autonomously locates and builds maps with visual sensors.In order to meet the requirements of outdoor navigation and map construction,a road SLAM algorithm based on binocular vision is proposed in this thesis,focusing on the improved algorithm of feature matching and scene reconstruction.The main research work of the thesis is as follows:Firstly,the research status of autonomous driving,binocular vision and SLAM technology is fully investigated,and the basic theory of binocular vision SLAM is introduced in detail.Secondly,in order to improve the low precision and slow speed of traditional image feature matching algorithm,such as the random sampling consistency algorithm,an improved algorithm based on grid motion statistics and block matching is proposed.The number of matches in the neighborhood grid of feature point is used as the matching criterion,and the matching features are selected by diamond search algorithm.The experimental results show that the improved feature matching algorithm can effectively improve the screening rate and speed of correct matching features.Thirdly,aiming at the problem that the semi-global stereo matching algorithm has fixed the penalty parameters and cannot properly deal with the parallax mutation of object edge,an adaptive penalty parameter algorithm based on edge is proposed to optimize the matching of edge pixels and continuous pixels.At the same time,according to the disparity estimated by the improved stereo matching algorithm,the 3D dense point cloud of road scene is reconstructed,and a multi-threshold region segmentation and recognition algorithm is proposed to eliminate the sky pixels and improve the reconstruction efficiency.The experimental results show that the improved stereo matching algorithm can effectively deal with the parallax mutation of object edge,and improve the estimation accuracy of the parallax.In addition,the object surface smoothness,scene density and reconstruction time of 3D dense point cloud can meet the outdoor requirements.At last,according to the requirement of autonomous driving vehicles,a road SLAM algorithm based on binocular vision is proposed.On the basis of ORB-SLAM2,the map construction thread is added.By combining the improved feature matching algorithm and the road scene reconstruction algorithm,the vehicle is positioned and the dense point cloud map of road is constructed.The KITTI dataset is used for the simulation experiments of location estimation and map construction,and the actual experiment is carried out with a corrected binocular camera.The experimental results show that the proposed road SLAM can estimate the motion trajectory of the camera in real time and construct a dense map of road,which verifies the feasibility of the algorithm in practice.