Modified Keyframe Selection Algorithm and Map Visualization Based on ORB-SLAM2

Recently, Simultaneous Localization and Mapping (SLAM) has becoming an important technology for mobile robot. Real time performance and localization accuracy are two main indicators for SLAM. Traditional SLAM algorithms (i.e. ORB-SLAM2) adopt interframe motion time based method, which depends on keyframes similarity detection to select keyframes. In contrast to the old keyframe selection method that generates excessive keyframes and sparser pointclouds. We propose a novel method called repeated motion detection (RMD) to modify the keyframe selection section. RMD can effectively detect and eliminate the redundant keyframes generated by the camera’s back and forth motion, and thus improve the real-time performance, mapping and localization accuracy of the algorithm. Furthermore, we use this modified algorithm to build the Octomap and 2D occupancy grid map and compare ours with classic SLAM algorithm ORB-SLAM2. The experiment results show that our method achieve better result in both real time performance and localization accuracy.