Fast Stereovision by Coherence Detection

In human stereovision, a vivid sensation of depth is created by the small relative displacements of objects in the retinal images of the left and right eye. Utilizing only two diierent views of a scene, the range of disparities which simple disparity units can estimate is severely limited by aliasing eeects. A new computational approach to stereo vision utilizes these aliasing eeects in a coherence detection scheme to allow the calculation of dense disparity maps over a large range of disparities with plausible biological hardware. The fast, non-iterative algorithm creates within a single network structure simultaneously a disparity map exhibiting hyperacuity, a veriication count of the disparity data and fuses the left and right input images to the cyclopean view of the scene.