Recovering 3-D translational motion and establishing stereo correspondence from binocular image flows

Stereo imagery with translational motion is analyzed. It is shown that the relative translational velocity between the camera platform and the objects can be computed by solving linear equations based on the measured flow fields of the left and right cameras, without point-to-point correspondence. In addition, stereo matching procedures based on the estimate translational velocity and the flow fields are presented. Preliminary results with synthetic data show that these techniques are quite robust in the presence of noise