The Application of a Particle Filter to Urban Ground Target Localization, Tracking, and Intercept

Abstract

Ground vehicle localization is a problem of significance in an urban setting given the recent global conflicts, security interests, and rapid growth of sensor networks. This task is often difficult due to the lack of information regarding a target vehicle's position, velocity, and destination. Field operatives can provide binary measurements of a target's presence in an area, and these measurements can be processed to obtain estimates of the target's location. A particle filter is more suitable for this application than a Kalman filter due to its ability to handle non-Gaussian distributions and non-differentiable measurement models, however it is computationally expensive. Suppose there is a mobile ground vehicle of known description but unknown position, velocity, or destination that is to be found, tracked, and intercepted by an unmanned aerial vehicle. The vehicle is known to be in an urban environment, and full knowledge of that environment (roads, obstacles, intersection constraints, and speed limits) is available. There are numerous issues of interest within this problem. A particle filter in an urban environment was developed to locate, track, and intercept a ground vehicle given soft binary measurements (measurements from human sources). Two particular issues are studied in this work: the effect of a sophisticated particle dynamic model on target localization and tracking, and the development of a real time path planning routine in the particle filter framework to enable target interception. The contributions of this work are threefold. First, the importance and impact of an accurate particle time update on target localization and tracking is validated. Secondly, a thorough investigation into the effect of particle spatial resolution in the presence of imperfect measurements is made that will prove valuable for future particle filter applications. Finally, the path planning routine offers reduced computational expense when compared to existing systems and lends itself to unmanned aerial vehicle implementation. Proper exploitation and implementation of the particle filter framework prove vital in the complete characterization of the urban tracking problem.