基于Kinect传感器的三维点云地图构建与优化

Building and Optimization of Kinect-based 3D Point Cloud Map

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。

For the problem of poor robustness and real-time performance of mobile robots in the process of 3D point map building, a method of 3D map building based on image feature points was proposed. First, feature point extraction and matching was performed on the RGB data collected by Kinect. The RANSAC algorithm was used to eliminate the false matching points, and the number of iterations for the registration algorithm is effectively reduced while ensuring the accuracy. Combining with Kinect depth data, the pose of the corresponding feature point pairs in 3D space was obtained. Finally, ICP algorithm was used to iteratively solve the rigid transformation matrix for accurate registration to obtain indoor 3D point cloud map. To suppress the pose drift caused by accumulated error in the process of 3D point cloud registration, the loop closure detection based on TORO graph-based algorithm was introduced. Experiments verifies the effectiveness of the proposed method.