Research And Implementation Of SLAM Algorithm Based On Binocular Vision

SLAM(Simultaneous localization and mapping)refers to the construction of environmental maps and self-positioning based on information acquired by sensors.Because of the low cost,low power consumption and rich information,visual SLAM has become one of the key technologies for mobile robot autonomous navigation.This paper mainly uses the binocular camera to study the visual SLAM.In this paper,aiming at the error of mobile robot simultaneous positioning and map construction,this paper proposes to use four images of adjacent frames of binocular camera for feature matching to improve the accuracy and robustness of mobile robot pose estimation and map construction.The main work of this paper is as follows:1.For the feature extraction and matching of feature points in binocular vision SLAM,this paper uses four pictures of two frames adjacent to the binocular camera to perform feature matching.If all four pictures can be matched successfully,it is considered to be a valid match.More accurate feature points.At the same time,the matching points are triangulated to eliminate the mismatch,so that more accurate feature points can be obtained.2.Aiming at the problem of accurately solving the pose of mobile robot in SLAM front end,this paper assigns different weights to different feature points in the initial pose estimation,which reduces the error.The obtained mobile robot pose is set as the initial value,and then the pose optimization and the feature points are jointly optimized by the method of graph optimization to obtain a more accurate solution.By using the algorithm,a more accurate pose estimation value can be obtained.3.Aiming at the real-time problem of binocular vision SLAM,this paper uses the sparsity of the matrix in the back-end optimization to accelerate the operation.After optimization,the feature points tend to converge to a fixed value.At this time,only the constraint relationship between poses is used to optimize the pose,and the optimization efficiency of the backend is improved.4.Aiming at how to carry out loop detection and optimization,this paper adopts the closed-loop detection algorithm based on visual word bag model to detect the similarity between the current key frame and the previous key frame,and judge whether there is loopback according to the similarity.If there is loopback in the detection,loopback detection is performed.The results are passed to the backend for optimization to reduce cumulative errors.Loopback detection with this method has higher efficiency and accuracy.Finally,summarize the full text and propose a prospect for further research.