An Improved Stereo Visual-Inertial SLAM Algorithm Based on Point-and-Line Features for Subterranean Environments
Qian Sun, Hao Wang, Wa Liu, Junjing Zou, Fang Ye, Yibing Li
Abstract
Simultaneous Localization and Mapping (SLAM) in GPS-denied, perceptually-degraded and complex subterranean environments is a challenging problem. A stereo visual-inertial SLAM algorithm for subterranean environments is proposed in this paper. Firstly, an image enhancement algorithm for SLAM is proposed to enhance low-light images of subterranean environments in real time. Secondly, line features, represented by utilizing Plücker coordinates and standard orthogonal, are added on the basis of existing point features. To reduce the computational burden and maintain real-time performance, an improved line feature matching and screening strategy is proposed by using the line feature geometric feature filter. Experimental results in EuRoC, OIVIO datasets and real subterranean environments show that the method proposed in this paper possesses high positioning accuracy and system robustness.