We propose a new model free approach to object tracking. It is based on the estimation of the location and scale parameter of a feature distribution by using asymptotic properties of order statistics. The basic distribution considered is the distribution of the coordinates of some kind of low level localized features, e.g. edges. A robust and real time tracking behavior is achieved by the algorithm's integrative character. The performance of the approach is demonstrated by a real world example.
展开▼
