Fast stereo visual odometry based on LK optical flow and ORB-SLAM2

Abstract

A stereo visual odometry algorithm based on the fusion of optical flow tracking and feature matching called LK-ORB-SLAM2 was proposed. In LK-ORB-SLAM2, the operation of optical flow tracking is introduced to adjust the intensive and time-consuming operation of feature matching. This requires solving a key issue: how to solve the problem of losing feature points during optical flow tracking. For this reason, an adaptive matching-frame insertion scheme is proposed to stop optical flow tracking in time and inserts matching-frames and detect new feature points at the right time to keep LK-ORB-SLAM2 running. The experiment on the KITTI and EuRoC data set showed that LK-ORB-SLAM2 reduced the average processing time per frame of ORB-SLAM2 by about 70%, with the change of less than 2% in its accuracy.