无人系统视觉SLAM技术发展现状简析

Visual simultaneous localization and mapping(vSLAM) plays a vital role in fully autonomous navigation and perception of unmanned systems in unknown environment. Following a brief introduction to the history and typical structure of vSLAM, a summary about state-of-the-art representatives, advantages and disadvantages of two front-end pose estimation methods(feature based vs. photometric based) and two back-end optimization methods(nonlinear filter vs. nonlinear optimization) is given. On this basis, relevant achievements of visual inertial SLAM(VI-SLAM) are classified into different categories according to their coupled types and back-end optimization methods. Furthermore, the similarities and differences of representative open-source vSLAM frameworks are compared and analyzed, alongside with the comparison of their performance under public datasets. Finally, challenges faced by current vSLAM are elaborated from the aspects of scenario generalization, advanced perception, dynamic adaptability and multi-sensor integration.Future development trends and directions of vSLAM are also discussed to serve as an useful guide for researchers.