Abstract
We propose a sequential Bayesian method for estimating the underlying motion governing a group of points by tracking the individual points separately. The proposed method can be applied to the problem of tracking an extended object or a group of objects moving in formation. To this end, a time series of the observed position of a point is decomposed into a combination of the underlying motions such as rotation and translation by a particle filter. The parameters of the underlying motions are estimated by cumulating the posterior distributions of the group of points. The proposed method can reduce the ambiguity in decomposing a series of the low dimensional observations into a series of the higher dimensional states without data association. Experiments with numerical data show the effectiveness of the proposed method.
