Skip to main content

Hierarchical models for visual shape and motion


Research objectives and content
Object tracking is an important and still difficult problem. For example surveillance applications require the tracking of intruders who might perform complex and erratic motions. The person might for example crawl, walk, run through a gate or jump over a fence. Moreover, once an object has been tracked, additional evaluation is needed on order to decide the exact nature of the motion and to react appropriately. These issues of tracking and interpretation are the focus of this research proposal. The main aim of this research is to build a system which handles object tracking and motion classification simultaneously. Until recently tracking and motion classification have been dealt with as two separate processes. However it is highly desirable to develop systems where classification feeds back into the perception of motion since perception and classification are inextricably bound together. The reasoning behind this approach is that the statistical models used for the anticipation of motion can potentially be adapted for classification.
Training content (objective, benefit and expected impact)
It is my objective to obtain the academic grade Doctor of Philosophy of the University of Oxford. This research will allow me to explore and contribute to the state-of-the-art of visual tracking. In addition I will reinforce my knowledge in the field of statistical pattern recognition and neural networks. This will be an excellent basis to successfully undertake further research in the field of computer vision in the future. Two aspects make this research particularly interesting. This research itself is the first attempt to bind tracking and motion classification into one model. Furthermore the results have the potential to significantly improve surveillance application, an area in which very few researchers work, but which is of increasing interest to the industry.
Links with industry / industrial relevance (22)
The Visual Dynamics Research group is involved in a EU-project called IMPROOFS (IMage PROcessing Operations for Forensic Support) in which the project partners work in close cooperation with the UK Forensic Science Service (FSS). The FSS will also be consulting me in the course of the project. In addition I have contacted the Automation Group of Siemens regarding motion classification for surveillance applications. As a result of this they have sent me an extensive data base of video sequences that will be used for the experiments.


Parks Road
OX1 3PJ Oxford
United Kingdom