Fusion of frequency and spatial domain information for motion analysis

This work presents an approach to the analysis of multiple motions in video, which combines frequency and spatial domain information in a new manner. The tasks of interest are finding the number of moving objects, velocity estimation, object tracking, and motion segmentation. We propose a novel, hybrid approach, which avoids problems of spatial domain methods, like sensitivity to illumination changes or great computational cost, but also uses spatial information for precise object localization. Experiments with synthetic and real sequences, show both the possibilities and limitations of this approach.