论文部分内容阅读
针对移动机器人在三维点云地图创建过程中存在鲁棒性和实时性不佳的问题,提出一种基于图像特征点的三维地图创建方法。首先,对Kinect采集得到的RGB数据进行特征点提取与匹配,并采用RANSAC算法对误匹配点进行剔除,在保证精度的同时,有效减少了配准算法的迭代次数,通过结合Kinect深度数据得到对应特征点对在三维空间中的位姿,最后采用ICP算法迭代求解刚体变换矩阵完成精确配准,得到室内真实场景下的三维点云地图。为抑制由三维点云配准过程中累积误差造成的位姿漂移,引入了基于TORO图优化算法的闭环检测机制,实验验证了所提方法的有效性。
Aiming at the problem that the mobile robot has poor robustness and real-time in the process of 3D point cloud map creation, a 3D map creation method based on image feature points is proposed. Firstly, the feature points of the RGB data collected by Kinect are extracted and matched, and the missing matching points are removed by using RANSAC algorithm. While ensuring the accuracy, the number of iterations of the registration algorithm is effectively reduced. By combining the Kinect depth data, Finally, using the ICP algorithm iteratively solves the rigid-body transformation matrix to complete the accurate registration, the 3D point cloud map in the real scene of the interior is obtained. In order to suppress the pose drift caused by the accumulated error in 3D point cloud registration, a closed-loop detection mechanism based on TORO graph optimization algorithm was introduced. The effectiveness of the proposed method was verified by experiments.