一种基于深度相机的机器人室内导航点云地图生成方法

摘要

点云地图是智能机器人自主导航的基础.文中提出一种基于深度相机的机器人室内导航点云地图生成方法,通过对图像特征的快速提取与匹配,实时估计相机位姿;综合考虑彩色图像的投影误差与深度图像的反投影误差,应用图优化算法对关键帧位姿与地图点进行联合优化;通过环路检测与地图优化降低估计误差累积的影响;利用估计得到的相机位姿将关键帧对应的图像点云进行拼接融合,形成表示三维空间场景结构的稠密点云地图.通过实验验证方法的有效性、精确性与实时性.%Point cloud map is the foundation for auto navigation of intelligent robots.A method is proposed to generate indoor point cloud maps for auto-navigation of robots,so the feature points are detected and the correspondences between consecutive color images to estimate real-time pose of the camera are found. The graph optimization algorithm is used to optimize the pose of newly added key frame and associated key frames and map points.Taking the errors of projection of map points and reconstruction of image pixels into consideration,this method eliminateds the effects of accumulate errors with the detection of loop closure and optimization of map,stitched and fused point clouds associated with key frames with camera pose estimated to get the indoor map of dense point clouds representing the scene structure of 3D space, which proves the effectiveness,accuracy and real-time capability with experiments.