Study On Simultaneous Localization And Mapping Methods Based On Binocular Vision

Simultaneous localization and mapping technology refers to autonomous localization and map building based on its own sensors without prior map information.With the development of intelligent technologies,simultaneous localization and mapping(SLAM)has become a key technology in the fields of unmanned vehicles,robots,augmented reality,virtual reality and so on.And it has received extensive attention.SLAM based on binocular vision has broad application prospects because of its simple sensor structure and rich information acquisition.This paper adopts the scheme of implementing real-time location and sparse points map construction by extracting image feature of the binocular image and feature matching.Then improvement and argumentation analysis of feature extraction,feature matching optimization,loop closing detection and so on were carried out in the paper.The main work includes:Designed binocular vision SLAM framework with front-end and back-end.The front-end mainly performs feature extraction and feature matching,camera pose calculation and map points construction,and the back-end is mainly responsible for loop closing and optimization to correct accumulated errors.The back-end optimization can be further divided into local optimization and global optimization.The purpose of local optimization is to minimize the re-projection error in the process of camera motion,and the global optimization is based on the loop closing detection results to introduce the trajectory loop closing constraint for global data optimization.This paper selects the BA map optimization algorithm with obvious advantages for map optimization.Improved ORB feature extraction algorithm and feature matching optimization scheme.This paper selects the efficient ORB feature and introduces the image pyramid based on the original ORB feature algorithm to improve the robustness of the ORB feature to the scale change.In order to improve the feature matching efficiency and eliminate feature mismatches,this paper introduces the quadtree index and PROSAC algorithm respectively.Experiments show that the above algorithm combination has obvious advantages in feature extraction and matching.Research on direct loop closing detection algorithm based on multi-Index hash table.Accurate loop closing detection is essential for building globally consistentcamera trajectories and maps,the commonly used loop closing detection algorithm based on bag of words is too dependent on the dictionary.This paper uses a loop closing detection algorithm based on multi-index hash table to directly use the feature points of image frames to detect loop closing directly.Comparing the experimental results,the loop detection algorithm not only outperforms the loop closing detection algorithm based on bag of words in the accurate loop closing detection,but also meets the real-time requirements.Finally,through the experimental analysis of the whole binocular vision SLAM system proposed in this paper,it can be concluded that it can accurately autonomous positioning and map construction.