Estimation of multiple, time-varying motions using time-frequency representations and moving-objects segmentation.

Abstract

We extend existing spatiotemporal approaches to handle time-varying motions estimation of multiple objects. It is shown that multiple, time-varying motions estimation is equivalent to the instantaneous frequency estimation of superpositioned FM sinusoids. Therefore, we apply established signal processing tools, such as time-frequency representations to show that for each time instant, the energy is concentrated along planes in the 3-D space: spatial frequencies-instantaneous frequency. Using fuzzy C-planes, we estimate indirectly the instantaneous velocities. Furthermore, adapting existing approaches to our problem, we attain the identification of the moving objects. The experimental results verify the effectiveness of our methodology.