Particle Filters and MAP Sequence Estimation for Vehicle Tracking

Efficient methods for estimating the maximum a posteriori (MAP) sequence of a Markov process have recently been developed for particle filters, which extend the Viterbi algorithm to continuous, non-linear processes. Vehicle tracking using an unmanned aircraft system (UAS) is one possible application where these methods can be used to make the association process more robust when similar vehicles or other objects are within the range of the sensors. This paper presents several particle filters of different complexity for vehicle tracking and their capability to perform MAP sequence estimation is evaluated using real sensor data. An efficient gradient-based post optimization method is also developed and evaluated. It is shown that the post optimization is a very good complement to the particle- based estimation due to its ability to efficiently compensate for lack of particles.