More details about the project  

Funded by the Data and Information Fusion Defence Technology Centre (DIF-DTF), the project is to develop a novel numerical-based methodology for real-time object detection and tracking. Classic applications of this project include radar and sonar, but the principles extend into computer vision, robotics and many other areas. We have aimed to push back the boundaries of current technology where many objects are present, detection probabilities are low and clutter rates are high. The methods use novel implementations of Monte Carlo Bayesian updating to carry out joint detection of number, characteristics and position of objects in cluttered environments.

Whereas classical methods can easily fail to operate, the new methods have been successfully evaluated under real-life data taken from a military submarine. The methods are also shown to have outperformed other classical methods in the cases when objects are travelling in very high speed (as high as 3 to 6 Mach) and manoeuvres. The output of this project will be integrated with those from other collaborators, including QinetiQ, Bristol University, and Imperial College, in other areas, including image tracking, to improve the overall demonstrable performance in advanced tracking techniques and research.

My future research in the group will be an extension of my current research. Also funded by the DIF-DTF, the proposed project will last for 2 to 3 years, starting from October 2006. Professor Simon Godsill and I will be responsible for developing a novel methodology, based on the knowledge and methodologies gained from the last phase, for real-time detection and tracking for clusters or groups of objects. The objective of this phase is more challenging because the objects to be tracked may no longer be travelling independently but in a coordinated format. Examples of these objects are coordinated flights of aircrafts and convoys of ground vehicles.

To tackle this problem, sophisticated interaction models for individuals within a group will be formulated, based upon pairwise spatio-temporal interaction models. Ultimately the methods will be developed in order to perform automatic convoy spotting where these convoys may have different intentions that may evolve with time. The required computational methodology will focus on multi-target Bayesian models using combinations of particle filter and Markov chain Monte Carlo methods. A particular methodological focus will be the scalability of the methods with group size. It has been advised that data from a commercial high range-accuracy radar, where individual targets result in multiple resolved detections, will be provided to help guide the research. Furthermore, in this project we also will study the characteristics of large groups, or ‘clouds?of objects, in which individuals cannot be reliably tracked, but the whole group can be tracked through conglomeration of the many returns made by individuals in the cloud over a potentially large spatial area. It is planned to extend the use of Poisson models for clouds to achieve this aim, which removes the necessity to consider data association explicitly, hence making the framework tractable.

top of page