Using colour features for video-based tracking of people in a multi-camera environment

This paper provides a brief overview of ongoing work towards the realisation of a multi-camera video-based person tracking system. The developing system is orientated towards a distributed architecture for sharing visual information between multiple cameras. A hierarchical method for modelling people in scenes using colour features is proposed, however, future development will also look at including other useful measures such as texture. The method uses a kernel-based classification implementation to track a base set of visual components which can then be grouped to form high-level object descriptors. Some results are shown for a Kernel Adatron implementation of the classifier.