无人机视觉SLAM环境感知发展研究

Research on Development of UAV Visual SLAM Environment Perception

摘要:

无人机在进行搜索救援等高级任务的时候，往往需要确定自己的位置和环境信息。仿照于人类通过视觉感知环境信息，视觉SLAM(visual simultaneous localization and mapping, VSLAM)是计算机视觉领域里面通过视觉传感器感知环境的信息并快速跟踪自身的位置和建立环境地图的一种前沿技术。本文首先阐述了VSLAM的重要组成部分：前端处理（特征点法和直接法）、数据关联、后端优化算法（滤波方法和优化方法）和建图；然后总结了一些在无人机上成功应用的典型VSLAM算法，以及在VSLAM发展的30多年的时间里涌现出许多出色的方案和研究机构；接着论述了当前用于无人机VSLAM发展的几个重点问题，多无人机协同的C-SLAM、深度学习和语义分割在SLAM中的应用、以视觉惯导为代表的多传感器融合SLAM；最后，对VSLAM方法进行总结，给出了未来的发展方向，希望对后续研究提供指导和帮助。

Abstract:

When UAV perform advanced tasks such as search and rescue, they often need to determine their own location and environmental information. Modeled on humans perceiving environmental information through vision, visual SLAM (visual simultaneous localization and mapping, VSLAM) is a cutting-edge technology in the field of computer vision that uses visual sensors to perceive environmental information and quickly track its own location and build environmental maps. This article first explains the important components of VSLAM: front-end processing (feature point method and direct method), data association, back-end optimization algorithms (filtering methods and optimization methods) and mapping; then summarizes some of the successful applications on UAVs Typical VSLAM algorithm, and many outstanding programs and research institutions that have emerged during the development of VSLAM for more than 30 years; Then it discusses several key issues currently used in the development of UAV VSLAM, the application of multi-UAV collaborative C-SLAM, deep learning and semantic segmentation in SLAM, and multi-sensor fusion SLAM represented by visual inertial navigation; Finally, the VSLAM method is summarized, and the future development direction is given, hoping to provide guidance and help for follow-up research.