Autonomous Inspection for Mobile Robot Based on Visual and Inertial SLAM

This paper introduces a position acquisition method and post-matching processing technique tailored for Simultaneous Localization and Mapping (SLAM) based on visual -inertial fusion. In visual-inertial SLAM systems, the position estimation accuracy is improved by trust region dogleg optimization algorithm combined with manifold optimization in response to the problem of the limited accuracy of Direct Linear Transform (DLT) in the initialization stage. An inlier set generation technique based on second-order spatial compatibility (SC2) is proposed to improve the problem of partial error matching which is unavoidable in feature matching. In the feature tracking stage, a number of compatible point pairs are selected as seed points. In the loop closing stage, points with higher confidence are selected as seed points based on the principal eigenvector of the adjacency matrix of the graph. The VI-SLAM system is deployed and integrated to the physical prototype of the inspection robot, and the inspection capability of the robot is verified in indoor and outdoor scenarios with different characteristics. The results show that the inspection robot system built in this paper can stably complete inspection tasks in multiple scenarios.