A SMC Sampler for Joint Tracking and Destination Estimation from Noisy Data

In this paper we describe a Sequential Monte Carlo (SMC) sampler that performs joint tracking and destination estimation of a target traveling along a known road network, as its journey progresses. The destination estimation is based on a simplistic model of driver intent, which assumes no prior knowledge of the history of visited destinations. The proposed algorithm is capable of refining the distribution of destinations that can be inferred from an incoming stream of position estimates. We compare the performance achieved by the proposed algorithm with a mainstay Particle Filter, demonstrating how the later suffers greatly from sample impoverishment, therefore necessitating an ever increasing number of particles as the number of possible destinations increases, while showcasing that the issue is significantly mitigated by the proposed SMC Sampler.