Motion detection using 3D-FFT spectrum

A novel motion detection technique for multiple objects from image sequences is proposed. The algorithm is based on directional filtering in the spatiotemporal frequency domain. By using the 3-D fast Fourier transform (FFT), because of its computational efficiency, the spectrum of an object moving at constant velocity lies in several parallel planes in the frequency domain. The object's velocity is estimated by detecting the orientation of those planes. This problem can be reduced to detecting the orientation of parallel lines in the plane intersecting those planes by using direction-tuned filters. Statistical estimation is used to eliminate the uncertainty which originates from nonorthogonality of the filters. Experimental results on computer-generated images and real-life video images are also described.<
>