快速室内视觉同步定位与建图研究

Research on fast visual simultaneous indoor localization and mapping

家居机器人技术一般应用视觉同步定位与建图(SLAM,Simultaneous Localization and Mapping)来实现定位与构建导航地图,如何实现视觉SLAM系统快速准确定位和构建丰富环境信息的地图已经成为视觉SLAM研究的热点问题。本文将光流法与关键点结合,加快视觉SLAM的数据处理速度,并添加稠密点云地图和八叉树地图构建线程来获取环境信息,实现一个较为优秀的视觉SLAM系统。在公开数据集上进行的定位实验表明,该视觉SLAM系统在绝对轨迹误差和相对位姿误差上与ORB-SLAM2系统保持基本一致,并且在其中几项数据中具有更小的误差结果,整体系统对图像的处理速度约为40 FPS(Frames Per Second),是ORB-SLAM2系统的1.4倍左右,说明该系统在提高系统速度的基础上保持了较高的准确度。

Home service robot technology generally uses visual simultaneous localization and mapping(SLAM)to achieve positioning and establish navigation map.How to realize the rapid and accurate positioning of visual SLAM system and build a map with rich environmental information has become a hot topic of visual SLAM research.Under this demand,this paper combines Optical Flow method with key points to speed up the data processing speed of visual SLAM,and adds dense point cloud map and octree map construction thread to obtain environment information,and realizes an excellent visual SLAM system.Experiments on open datasets show that the visual SLAM system in this paper is basically consistent with ORB-SLAM2 system in absolute trajectory error and relative pose error,and has smaller error results in several data,and the image processing speed of the whole system is about 1.4 times of that of ORB-SLAM2 system,which shows that the system maintains high accuracy on the basis of improving the system speed.