Semi-Direct Simultaneous Localization And Mapping With Stereo Camera

Recently,Augmented reality and autonomous driving have attracted more and more attention,which promotes the development of simultaneous localization and mapping with high precision and robustness.Scholars have successively proposed high-precision simultane-ous localization and mapping with stere camrea based on the research of existing monocular simultaneous localization and mapping technologies,such as ORB-SLAM2[1]?Stereo DSO[2].In this paper,the natural advantages of stereo camera compared with the monocular simultaneous localization and mapping are fully utillized.A high precision visual odometry w ith stereo camera based on direct method combining feature point method is proposed.Combining word bag method for loop detection and relocation,a simultaneous localizat.ion and mapping with stere camrea sy stem based on direct method and feature point method is proposed and implemented.The following are the main contributions of this paper:1)Based on the framework of Direct Sparse Odometry,a semi-direct visual odometry with stereo camera is proposed.The information brought by stereo camrea is used to solve the problem of scale uncertainty and improve the efficiency and success rate of initialization and the accuracy of camera poses and map points.At the same time,we combine the advantages of feature stability of feature point method and add feature re-projection error between key frames in local bundle adjustment to improve the robustness of the system.Finally,the visual odeometry with stereo camera is up t.o the state-of-the-art level.2)On the basis of the visual odometry with stereo camera,combined with the method of visual w-ord bag,the loop detection and relocation are added to the system,which can effectively solve the problem of loop closure and eliminate accumulated errors,and forme a complete simultaneous localization and mapping with stere camera system.