Novel methods for multi-target tracking with applications in sensor registration and fusion

Abstract

Maintaining surveillance over vast volumes of space is an increasingly important capability for the defence industry. A clearer and more accurate picture of a surveillance region could be obtained through sensor fusion between a network of sensors. However, this accurate picture is dependent on the sensor registration being resolved. Any inaccuracies in sensor location or orientation can manifest themselves into the sensor measurements that are used in the fusion process, and lead to poor target tracking performance. Solutions previously proposed in the literature for the sensor registration problem have been based on a number of assumptions that do not always hold in practice, such as having a synchronous network and having small, static registration errors. This thesis will propose a number of solutions to resolving the sensor registration and sensor fusion problems jointly in an efficient manner. The assumptions made in previous works will be loosened or removed, making the solutions more applicable to problems that we are likely to see in practice. The proposed methods will be applied to both simulated data, and a segment of data taken from a live trial in the field.