Morphing as a tool for motion modelling

Abstract

This study merges the methodologies of computer graphics and image processing by using morphing techniques in motion modelling problems. The proposed method morphs the final frame of a sequence of motion frames from the initial frame for modelling the motion of a smoothly changing video. First a set of control points are extracted in the initial and final frames of the motion sequence. For this purpose, the moving pixels are identified and clustered. The cluster centers of the moving pixels are labelled as control points. The number of control points and the motion detection thresholds are identified during the morphing by minimising the sum of the square error between the actual and generated image sequence. The control points are morphed for finding the trajectories of the motion. Then, the rest of the points are morphed by using the Gaussian distribution centred at each control point. The variance of each Gaussian distribution is estimated using the intermediate frames. The experimental results indicate very promising results.