Motion analysis: model selection and motion segmentation

Abstract

A new model selection criterion based on physical characteristics of underlying motion models is proposed. The proposed criterion is then incorporated in a robust motion segmentation scheme, which is based upon robust least K-th order statistical model fitting. The proposed model criterion has been compared with many other competing techniques and is shown to be more suitable for the motion segmentation task. The motion segmentation algorithm has been tested (and shown to be successful) on a number of synthetic and real image sequences.