基于XTION的机器人室内三维地图实时创建

摘要

针对室内复杂环境中移动机器人三维地图创建问题,基于RGB-D相机XTION搭建机器人地图创建系统,提出一种利用视觉里程计和贝叶斯闭环检测实时创建三维地图的方法.在机器人操作系统(ROS)框架内,首先由移动机器人携带XTION获取环境三维图像经WIFI上传上位机,然后由上位机软件使用SURF算法对RGB-D相机图像进行特征点提取与匹配,接着采用视觉里程计进行机器人位姿估计,同时利用贝叶斯闭环检测消除累积误差,最后使用g2o方法对三维点云图进行优化并创建完整的三维地图.经室内环境实验验证,机器人地图创建系统创建了完整的室内环境三维地图,所描述室内物体轮廓比较清晰并与真实场景相一致,机器人位姿估计比较精确,验证了利用视觉里程计和贝叶斯闭环检测实时创建三维地图的可行性.%On the indoor 3D map building for mobile robot, this paper, based on the robot map building system of XTION RGB-D camera, proposes a method of real time 3D map building with visual odometry and Bayesian closed loop detection, which requires the following process within the framework of the robot operating system (ROS): Firstly, the mobile robot with the XTION obtains the three-dimensional image of the environment, then uploads it to host computer through WIFI;Secondly, the host computer uses SURF algorithm to extract and match the feature points of RGB-D image with PC software;Thirdly, use the visual odometry to carry out the pose estimation of the robot and Bayesian closed loop detection to eliminate cumulative error;Lastly, apply g2o method to optimize the 3D point cloud image then build a complete 3D map.Under an indoor experiment, the robot map building system creates a complete three-dimensional map of the indoor environment, which clearly describes the real scene and contours of indoor objects, what's more, the pose estimation of robot is proved accurate.Thus, the experiment results support the method of using visual odometry and Bayesian closed loop detection to create real time 3D map.