In several applications of engineering, medicine and especially robotics, one often encounters the need to perform registration. In a typical registration problem, the spatial transformation between the geometric model of the object-of-interest and point measurements of the object's surface, needs to be estimated. In most applications, the point clouds obtained from sensors such as LIDAR, Kinect, feature rich stereo-images, etc, contain hundreds of points. Several methods have been developed to perform registration when dense point measurements are obtained. However, these methods do not perform well when only a small number of point measurements are available, and hence in this work we develop a method for robust sparse point registration (SPR).
展开▼