Monocular Semantic Visual SLAM Based On Point-line Fusion

SLAM(Simultaneous Localization and Mapping)technology is widely used in intelligent mobile devices,which can effectively provide intelligent mobile devices with their motion track,current pose and position information.At present,visual SLAM algorithms mainly use point features as constraints between frames to calculate pose.In complex dynamic low-texture scenes lacking sufficient point features,such as stairways and roads,the accuracy of pose calculation results is low.However,there are obvious line features and stable semantic information in this scenario.Therefore,this paper proposes a method combining point features,line features and semantic information to calculate pose between frames.Its main content and innovation are as follows:(1)A visual SLAM algorithm based on point-line fusion is proposed for low-texture scenes.Based on the framework of the classical ALGORITHM ORB-SLAM2,this algorithm uses THE ORB feature point detection algorithm to extract point features and the LSD detection algorithm to extract line features.The minimum reprojection error function of point features and line features is constructed to realize the fusion of point features and line features,which improves the robustness of visual SLAM algorithm.(2)Aiming at the problem of poor accuracy of pose estimation of SLAM system in dynamic environment,a semantic SLAM algorithm in dynamic environment is proposed.In this algorithm,the motion consistency detection algorithm and semantic segmentation algorithm are used to determine and eliminate the feature information of the dynamic object.Moreover,the medium and long term pose constraint conditions are established by using the semantic information and the drift correction is carried out by continuous tracking.So the influence of the dynamic object on pose estimation is reduced.(3)Based on the analysis of point and line features and semantic information in complex dynamic low-texture scenes,the projection error function of point and line features and semantic information is proposed,and the optimal pose is confirmed by minimizing the error function.Then,a monocular semantic visual SLAM algorithm based on point and line fusion is constructed.Compared with the traditional ORB＿SLAM2 algorithm based on point features,this algorithm improves the accuracy of pose calculation.