Determining stereo correspondences and egomotion from a sequence of stereo images

A trinocular stereo vision algorithm which uses motion information is introduced. This algorithm predicts current stereo matches from previous ones using a motion estimate. The cooperation between stereo and motion is used to reduce the computational load of stereo matching. Experiments with real data are presented to demonstrate the stereo and motion cooperation. In each camera, the optical flow is provided by a token tracker. It appears that the egomotion can be estimated using the optical flow from only one camera. However, more reliable motion estimates can be expected by combining the tracking results