Motion analysis using the 3-D Gabor transform

It has long been known that motion in a sequence of images has a characteristic signature in the spatiotemporal-frequency (Fourier) domain. For sequences which involve multiple objects in motion, however the global nature of the Fourier transform makes it unsuitable as a mechanism for motion analysis. In general, the signatures of the different motions cannot be resolved. Local frequency representations provide a means to address this difficulty. In the past, both the use of windowed Fourier analysis and the Wigner distribution have been proposed for this task. We consider the application of this 3-D Gabor (1946) transform to motion analysis.