Robust Particle Filter for Urban Radar Tracking using Terrain Blockage and Specular Multipath Detections

A real-time target state estimator is presented for the purpose of tracking a vehicle in a space bounded by linear position constraints, i.e., buildings described as polygons. The tracker utilizes measurements from polar-coordinate radar detections, from line-of-sight or building-reflected, single-bounce multipath and considers view blockage and shadow regions when the target is not detected. The method uses particle filtering for the predicted and posterior target state density and develops a Bayesian multipath measurement theory that drives the development of the feasible particle filter algorithm and the nearly-optimal importance sampling scheme that nearly eliminates the probability of particle impoverishment. Simulations show that the method effectively tracks the target with a low probability of impoverishment and demonstrate that the multipath particle filter with terrain awareness is capable of providing significantly better tracking accuracy compared to freespace tracking in geometrically favorable radar/target positions.