Tracking Skin-Colored Objects in Real-Time

Abstract

We present a methodology for tracking multiple skin-colored objects in a monocular image sequence. The proposed approach encompasses a collection of techniques that allow the modeling, detection and temporal association of skincolored objects across image sequences. A non-parametric model of skin color is employed. Skin-colored objects are detected with a Bayesian classifier that is bootstrapped with a small set of training data and refined through an off-line iterative training procedure. By using on-line adaptation of skin-color probabilities the classifier is able to cope with considerable illumination changes. Tracking over time is achieved by a novel technique that can handle multiple objects simultaneously. Tracked objects may move in complex trajectories, occlude each other in the field of view of a possibly moving camera and vary in number over time. A prototype implementation of the developed system operates on 320x240 live video in real time (28Hz), running on a conventional Pentium IV processor. Representative experimental results from the application of this prototype to image sequences are also presented. 1.