Combined Unscented Kalman and Particle Filtering for Tracking Closely Spaced Objects

Tracking closely spaced objects with resolution limited sensors is a difficult problem. One way to address this issue is to track these targets individually, and employ relatively complex data association approaches as a means of pairing detections and tracks. The algorithm outlined in this paper takes a different approach, and instead estimates the group velocity using an unscented Kalman filter (UKF). The UKF state estimate is then employed within a particle filter, which estimates the distribution of objects within the group. It is shown that this approach can be very effective, especially for groups of irregularly spaced objects