An evaluation of spatial mapping of indoor environment based on point cloud registration using Kinect sensor

Registration of 3D pointclouds obtained using depth sensor has wide range of applications in robotics. Many different rigid 3D registration algorithms have been proposed in literature, such as Principal Component Analysis, Singular value decomposition, iterative closest point (ICP) and its variants. The ICP is widely used algorithm for registration of point clouds. It is accurate and fast for point cloud registration. In this work, a performance evaluation of point-to-point based ICP algorithm, integration of point-to-point with random sampling and point-to-plane based ICP algorithm by using Microsoft Kinect camera is conducted. Low-cost Microsoft Kinect sensor provides a feasible and economical solution for such point cloud generation. Root mean square error (RMSE) value is taken as the measurement of precision of cloud registration. RMSE value is obtained from the Euclidean distance between corresponding point pairs in both point-clouds, used for the ICP registration. The ICP algorithm always converges monotonically to the nearest local minimum of a mean square distance metric. The results show that the convergence rate is fast during initial iterations.