Motion Estimation of Object based on Three-Dimensional Correspondence
Katsumi HAMA, Yukinori Kakazu · Journal of the Japan Society for Precision Engineering · 1993
This paper describes an approach for estimating motion parameters of a rigid body utilizing three-dimensional (3-D) correspondence. 3-D positional information of object feature points, which is available for object representation and robot vision, is extracted by stereo matching at each instant. The correspondence between two 3-D point sets at sequential time instances is needed to find motion parameters, the best one is established using Hopfield neural network. On the basis of acquired information, two methods for motion estimation are proposed (1) in an incremental manner under a condition for smoothness of motion, and (2) in an iterative operation according to a Daugman-type linear network architecture. The results of applying these schema to a sequence of stereo images are presented.