Detecting and Segmenting Motions by Matching Flows

The motion of an articulated object can be very complex but the motions of its rigid parts are usually simple. Here a method is proposed to classify and segment the motion of an articulated object into simple motions in temporal and spatial domains. Simple motions are represented by the corresponding motion flow patterns. Both the motion detection and segmentation are accomplished by matching the flow patterns of known simple motions with image flow. Matching is done by comparing the parametric flows of motion patterns with the normal flows of a target image. The approach is demonstrated on several image sequences of moving arms