多视角几何Rao-Blackwellised SLAM 算法

SLAM method with multiple vision geometric and Rao-Blackwellised particle filtering

低成本传感器、高精度定位是机器人同步定位与地图构建的热点研究问题。采用单视觉传感器来获取多个图像视觉信息,通过因式分解求解多幅图像的空间关系;利用多视角几何理论获得环境深度信息。机器人的同步定位与地图构建是通过对观测模型进行一阶泰勒近似,以及Rao-Blackwellised粒子滤波迭代进行的。在MT-R移动机器人研究平台进行实验,实验结果表明所提出的方法在定位精度指标优于经典的EKF-SLAM方法,并且只需要单个摄像头。

Low price sensor and high precise localization is a hot research issue. In this paper, only single camera is used to catch multiple images. The relation between images is gotten by factorization method. From the multiple geometric calculating, the deep on the environments is gotten. The robot simultaneous localization and mapping is performed iteratively by the Rao-Blackwellisated filter. The observation model is approximated with first-order Taylor. The experiments are performed with MT-R robot plant. The results show that the localization precise of robot with only single camera is more than the classic method.