COMPUTING 2-D MOTION FIELD WITH MULTIRESOLUTION IMAGES AND COOPERATION OF GRADIENT-BASED AND MATCHING-BASED SCHEMES

A new approach is presented for the detection and computation of a two -dimensional motion field in image sequences. This computational model has a multi-channel motion detector and an optimal motion selector. I n the motion detector, each channel has an inherent spatial resolution . The detector computes a two-dimensional motion field by the gradient -based method in parallel. The motion selector compares those candidat es of the motion field by a correlation value of the intensity pattern s hierarchically arranged from low to high resolution. It then determi nes the most probable motion for each image point. Experimental result s are shown for synthetic images. This model can detect more reliable motion fields than the conventional one-channel model.