基于特征点法和直接法VSLAM的研究

Research of feature-based and direct methods VSLAM

基于视觉的同时定位和建图(VSLAM)分为前端和后端,前端包括视觉里程计和回环检测,后端包括后端优化和建图。按照估计相机运动的不同方式,将VSLAM分为特征点法和直接法,首先从这两个方面对前端进行综述,阐述其中的关键技术和最新的研究进展,对比分析不同方法的优缺点;然后详细分析优化后端与滤波器后端的区别,进一步对多个开源代码进行比较研究,分析它们的优劣势和适用场合;再讨论深度学习、语义地图和多机器人在VSLAM领域的研究进展,以及相关技术与VSLAM的结合方式及前景;最后对VSLAM的未来进行展望。

VSLAM is divided into front-end and back-end.The front-end includes visual odometry and loop detection,and the back-end includes back-end optimization and mapping.This paper divided VSLAM into feature-based method and direct method according to different ways of estimating camera motion.Firstly,it summarized the front-end from these two aspects,elaborated the key technologies and the latest research progress,compared and analyzed the different methods.Then,it analyzed the differences between the optimize back-end and the filter back-end in detail,and compared the advantages and disadvantages of several open source codes and their applicable occasions.Further,it introduced the research progress of deep learning,semantic mapping and multi-robots in VSLAM,and discussed the combination of related technologies with VSLAM and its prospects.Finally,it prospected the future of VSLAM.