Research And Implementation Of Monocular Simultaneous Localization And Sparse Mapping

Simultaneous Localization and Mapping(SLAM)is considered to be the key technology to realize the autonomous positioning and navigation of mobile robots,and it is also a hot research technology in the field of unmanned,augmented reality and intelligent home.Monocular sensor is large favored because it has a lot of advantages such as low price,low power consumption,easy installation and use,and much information accessed.Monocular SLAM has become a hot topic in domestic and foreign research,which is of great significance to realize the autonomy and intelligence of mobile robot.The main work and innovation of this dissertation are as follows:First of all,this dissertation chooses the best of the state-of-the-art technologies by comparing the various implementations of the key technologies such as image feature extraction,map initialization,key frame selection,structure from motion,loop closure,and graph optimization in the current monocular SLAM of the program to achieve a SLAM system.Compared with the existing PTAM(Parallel Tracking and Mapping)system and the LSD-SLAM(Large-Scale Direct Monocular SLAM)system,this monocular SLAM system has improved performance in key frame track accuracy and robustness.Also it has automatic initialization and relocation.Secondly,in the newly constructed SLAM system,the "reverse replay" method is proposed to solve the problem of the loss of the key frame and map point between tracking loss and relocation,which can better use the input image information to enhance the integrity of the entire map.Finally,a comparative analysis experiment is designed for feature extraction algorithm,key frame selection method and closed-loop detection.It is proved that the ORB(Oriented FAST and Rotated BRIEF)feature extraction algorithm has great advantages in execution speed and the keyframe selection strategy of fusion of visual content change and time domain is better and the closed-loop detection method is valid and effective.This SLAM system,PTAM system and LSD-SLAM system are used to compare the mean square error of the key frame in the indoor data set.The accuracy of the SLAM system is better than that of the LSD-SLAM system and a little beter than the PTAM system.And the feasibility of this SLAM system implemented in this dissertation and the effectiveness of the "reverse replay" method are verified in the outdoor environment.