A robust elliptical head tracker

Abstract

A method is presented for robust head tracking in highly cluttered environments. The head's projection onto a 3D depth map of the scene is modelled as an ellipse whose position and size are constantly updated by a local search. Ellipse detection is performed on depth discontinuities using Chamfer Distance. The resultant system is a real-time, illumination and head-rotation invariant robust head tracker. The approach can be easily generalized further to perform tracking or detection of objects of different shapes e.g. hands.