Estimation of General Rigid Body Motion From a Long Sequence of Images

In estimating the 3-D rigid body motion and structure from time-varying images, most of previous approaches which exploit a large number of frames assume that the rotation, and the translation in some case, are constant. For a long sequence of images, this assumption in general is not valid. In this paper, we propose a new state estimation formulation for the general motion in which the 3-D translation and rotation are modeled as the polynomials of arbitrary order. Extended Kalman filter is used to find the estimates recursively from noisy images. A number of simulations including the Monte Carlo analysis are conducted to illustrate the performance of the proposed formulation.