I have started my PhD in Computer vision under the supervision of Prof. David Hogg in September 2006. I'm currently studying methods to improve the uncertainties in visual data. This is an extension to my MSc project entitled "Visual Signature for Large Scale Tracking". My PhD is sponsored by ORS and the School of Computing at the university.

Despite intensive research aiming towards universal tracking with “one object per track and one track per object” [Stauffer, 2005], such a tracker is not yet available if a single viewpoint is used. Understanding the expected events and scenarios is one way to assist in resolving conflicting and ambiguous observations.

The first scenario under investigation is studying the bicycle theft detection. The scenario where a person leaves an object (typically locked) within a parking /depositing area, and picks it up sometime later, presents a rich constrained scenario that may be observed by a single CCTV camera. However, uncertain observations from available tracking methods are often insufficient to recognize the actual events in isolation. The research accomplished up till the moment, proposes a method to connect people and objects, and decide on the sequence of drop-off and pick-up events. A link can then be inferred between the person dropping off an object, and the person picking up the same object later. Clothes colouring was utilized to compare these two individuals, and raise a theft warning when they look different.