An Efficient Framework for Particle Filter-Based Urban Target Tracking
Siddhartha S. Mehta, Pia E. K. Berg-Yuen, D. Waden, Eduardo L. Pasiliao, J. Willard Curtis · 2012
In this paper, we present an application of particle filter to target state estimation in an urban environment. The objective is to improve the estimation efficiency by 1) developing an accurate and comprehensive representation of the target task-space and 2) incorporating measurements from the readily available soft information sensors (SIS), e.g., verbal cues from a human observer along with the conventional sensor measurements. In addition, the sensor observation uncertainties are fused into the particle weight update law to inject robustness with respect to the classification type I and type II errors. Enhanced task-space representation is obtained by developing tools to exploit the open source map repositories as well as the a priori known urban terrain attributes to generate a sophisticated urban map. The target is modeled as a discrete-time nonlinear unicycle system with input velocity constraints and assumed to comply with the organized traffic rules. Extensive simulation results are presented that demonstrate improved estimation efficiency in the sense of mean squared error (MSE) and position variance using the proposed framework.