Multicamera Motion Estimation for High-Accuracy 3D Reconstruction

In this article, we approach the problem of accurately estimating the motion of a multi-ocular camera system from the analysis of point-like features extracted from the acquired images. We propose a method based on optimal 3D data fusion, and compare it with an approach based on projective constraints. We also assess the critical problem of the accurate feature localization on the image planes. The performance of the proposed solutions is here evaluated through simulation tests and through experiments conducted on real scenes using three standard TV-resolution CCD cameras.