Video Phase-Locked Loops in Gait Recognition

In the perception of gaits, the relative timing of the individual motions in a gait is critical. When events occur periodically, as they do in a gait, then relative timing is equivalent to phase. Therefore, to perceive gaits requires frequency entrainment (finding the fundamental frequency of the gait) and phase locking (finding the phase relationships among the components of the gait). In this paper, we present an innovation in motion recognition we call the video phase-locked loop (PLL). While the innovation is simple, the resulting system is uniquely well suited to recognizing gaits because it performs both frequency entrainment and phase locking. A video PLL produces a pattern of phasors (rotating phase vectors) over a region in an image that encapsulates the relative timing of motion in the images. We show how the Procrustes method of statistical shape analysis can be used to compute means over a set of phasor patterns and distances between pairs of patterns. They combine to form the basis of a gait recognition system: the means provide exemplars, while the distance measure allows matching of patterns to exemplars. We provide a set of examples that demonstrates the operation of video PLLs and the application of Procrustes shape analysis to the resulting phasor configurations.