Geometrical analysis of motion parameters from two sets of range image data

It is acknowledged in the literature that motion estimation is a fundamental problem in machine vision and robotics as it offers the potential for a large number of industrial and commercial applications to be developed. We present a novel approach to calibrate motion parameters based on the explicit use of distance between feature points and angle information. We extend Chasles´ theory by providing a closed form expression for the pole of a non-pure translational planar displacement and the screw axis of a non-pure translational spatial displacement through analysing the geometrical properties of correspondence vectors which have been synthesised into a single coordinate frame. We then propose two novel algorithms to estimate 2D and 3D motion parameters based on the explicit use of rigid constraints. For a comparative study of performance, we also implemented another well known motion estimation algorithm based on the constraint least squares method. Experimental results based on both synthetic and real range images have shown that the proposed algorithm is in general superior or similar to the constraint least squares based algorithm