Stereo Visual SLAM Using Bag of Point and Line Word Pairs

摘要

The traditional point-based SLAM algorithm performs poorly due to light changing, low-texture and highly similar scenes, while line segment features can better describe the structural information of the environment. For this problem, a new stereo visual SLAM system based on point and line features is proposed. The Jacobian matrix of the new optimization target combined with point and line features is derived in detail. At the same time, DBoW is extended with line features and the concept of point and line word pairs is proposed. The co-occurrence information and spatial proximity of point and line features are considered in loop closure detection. Experimental results on EuRoC and self-built datasets demonstrate that the proposed method outperforms ORB-SLAM2, which can reduce the localization error in both indoor and outdoor environments and improve the precision and recall of the loop closure detection.