Extended smoothing joint data association for multi-target tracking in cluttered environments

Abstract

In heavily cluttered environments, it is difficult to estimate the uncertain motion of an unknown number of targets with low detection probabilities. In particular, for tracking multiple targets, standard multi-target data association algorithms such as joint integrated probabilistic data association (JIPDA), face complexity and severely limited applicability due to a combinatorially increasing number of possible measurement-to-track associations. Smoothers refine the target estimates based on future scan information. However, in this complex surveillance scenario, existing smoothing algorithms often fail to track the true target trajectories. To overcome such difficulties, this study proposes a new smoothing joint measurement-to-track association algorithm called fixed-interval smoothing JIPDA for tracking extended target trajectories (FIsJIPDA). The algorithm employs two independent JIPDA filters: forward JIPDA (fJIPDA) and backward JIPDA (bJIPDA). fJIPDA tracks the target state forward in time and is computed after the smoothing is achieved. bJIPDA estimates the target state in the backward time sequence. The numerical simulation is performed in a heavily populated cluttered environment with low target-detection probabilities. The results show better target trajectory accuracy and false-track discrimination performance of FIsJIPDA compared with that of existing algorithms for tracking multiple extended targets.