Robust Translational Motion Parameters of Superposition-Type Planar Compound-Like Eye by the Parallel Binocular

摘要

This study develops a superposition-type planar compound-like eye (SPCE) which mimics the structure of a biological compound eye and enables self-motion estimation in computer vision applications. In the proposed approach, the conventional parallel binocular vision system is extended in both the horizontal and the vertical directions to create a planar array of charge coupled device (CCD) cameras. The configuration of the resulting SPCE can be regarded as a two-dimensional arrangement of single-row SPCE structures and side-by-side single-column SPCE structures. By stacking the image matrices and flow vectors of the individual SPCE structure, a generic model is constructed which enables the ego-motion of the SPCE configuration to be recovered using a conventional least squares estimation technique. The simulation results show that the proposed SPCE scheme achieves an excellent self-motion recovery performance even when the target object is subject to significant noise or when the CCD cameras within the SPCE configuration are displaced slightly such that they no longer have a completely uniform distribution. Furthermore, the results indicate that the relative error of the ego-translational motion estimates converges toward zero as the number or density of the CCD cameras in the SPCE configuration increases.