Improvement and Experimental Evaluation Based on ORB-SLAM-VI Algorithm

Abstract

With the development of science and technology in recent years, SLAM technology as the intersection of robotics and computer vision has also made great progress. Based on the simultaneous localization and mapping technology of visual-inertial (ORB-SLAM-VI) tight coupling feature point information, the number of feature points that can be extracted by robots in actual scenes or low-texture environments is small, and the application scenarios are not rich enough. As well as the problem that the algorithm initialization speed is not fast enough, a tightly coupled SLAM system based on the step-by-step joint initialization of point characteristics and IMU information is proposed. The improved algorithm has been experimentally proved on the EuRoC dataset sequence to initialize the system more quickly and stably. The time is shortened by at least 2 seconds; experiments on 11 different sequences of the dataset show that compared with the VINS-Fusion and ORB-SLAM-VI algorithms, the improved algorithm is less robust for localization in stereo-IMU scenarios, but the accuracy of pose tracking and positioning is the highest.