Vision SLAM-based UAV obstacle avoidance system.

The application of simultaneous localization and mapping (SLAM) technology enables Unmanned Aerial Vehicles (UAVs) to autonomously perform obstacle avoidance in unknown environments. In this context, this review introduces various frequently-used visual SLAM(V-SLAM) methods, including original mono SLAM, oriented fast and rotated brief (ORB)-SLAM, semi-direct monocular visual odometry (SVO)-SLAM, direct sparse odometry (DSO)-SLAM, etc., and categorizes them According to the front-end approach and the back-end approach, and also compares the robustness of these algorithms in different environments, whether they support loop detection and global optimization. The application and development of visual SLAM techniques on UAVs in recent years are also reviewed, as well as the outlook and predictions for the future of the V-SLAM field, including research on multi-UAV collaboration, application of deep learning, and semantic SLAM. In addition, algorithms related to obstacle avoidance are introduced and compared, including the artificial potential field method, rapidly-exploring random trees (RRT) algorithm and A-star (A*) algorithm, which are frequently used in the related research of obstacle avoidance, and their parameters such as determinism, global optimal path guarantee, and computational complexity are compared. [ABSTRACT FROM AUTHOR]