Determining 3-D Translational Motion by the Parallel Binocular

Estimating 3D motion parameters from a sequence of images is of crucial importance in many image analysis applications, ranging from robot vehicle guidance to obstacle avoidance, target tracking, the reconstruction of 3D models. In this paper, a novel algorithm to recover parameters for translational motion using such a parallel binocular system is presented. This approach overcomes the difficulty of matrix singularity happened in binocular. In order to fit into application requirements, a close-form solution is also derived. This solution owns some important features, such as no matrix manipulation, no danger of matrix singularity, easy to further analyze and apply. To validate this close-form solution, extensive experiments were conducted. Simulations in a perturbed environment are also performed to study the effect of possible image uncertainty. Especially during in perturbed environment, the issues that the errors in the Z-axis direction are considerably higher than those in either the X- or the Y-axes directions, and increase markedly as the level of perturbation increases are investigated.