摘要
在SLAM领域中,为了克服稀疏特征地图不能提供详尽环境信息的缺点,从观测信息的物理意义出发,提出了全局观测地图模型.其基本思想是在稀疏特征地图中嵌入全局密集地图信息,采用位移准则、特征准则和传感器量程准则提取必要的观测信息,然后对观测信息进行去噪、转换,接着根据观测信息的物理意义和机器人位姿估计的不确定性获取环境的全局密集地图,可视化后得到环境的二值地图、灰度地图或颜色地图.将全局观测地图模型与EKF-SLAM算法相结合,提出了GOE-SLAM算法,采用Car Park Dataset对GOE-SLAM进行了实验验证,结果表明GOE-SLAM生成了可信的密集地图,并且GOE-SLAM的计算复杂度与EKF-SLAM相当.
To solve the shortcoming of SLAM （simultaneous localization and mapping） that the sparse feature map fails to provide full information about the environment, global observation map model （GOMM） is proposed according to physical meanings of the observations. In GOMM, global dense map information is embedded into sparse feature map, and necessary observations are selected with displacement rule, feature rule, and sensory limit rule. After that, the selected observations are denoised and transformed. Then, global dense map of the environment is built according to physical meanings of the observations and uncertainty of the robotic pose estimation. Monochrome map, gray scale map, or color map of the environment is obtained after visualization. By combining GOMM with EKF-SLAM （extended Kalman filter SLAM）, an algorithm named GOE-SLAM （global observation EKF-SLAM） is put forward. Experiments with ＂Car Park Dataset＂ are carried out to evaluate the performance of GOE-SLAM. Experimental results indicate that a reliable dense map is built with GOE-SLAM, and the computational complexity of GOE-SLAM is nearly equal to that of EKF-SLAM.
出处
《机器人》
EI
CSCD
北大核心
2010年第5期647-654,共8页
Robot
基金
地面移动机器人大范围导航技术研究(预研基金)
