Abstract: In multi-target tracking, the real-time performance, state-estimates accuracy, and track continuity are affected by clutter, missed detection, and closely spaced targets. To solve these problems, an improved sequential Monte Carlo implementation (SMC) of the probability hypothesis density (PHD) filter is proposed. First, based on double one-to-one principles, particle labeling approach and weight redistribution scheme for particle cloud are proposed to shield against the negative effects of clutter in high prior density region and the detection uncertainty on the estimation. Second, the multi-estimate extraction is converted into multiple single-estimate extractions, which can provide the identity of the individual target; thus, explicit track maintenance can be obtained. Finally, a novel resampling scheme is proposed to reduce the effects of closely spaced targets on individual posterior information. The results of numerical experiments demonstrate that the proposed approach can achieve explicit track continuity and better performance compared to the basic SMC-PHD filter, in terms of faster processing speed and superior estimation accuracy.

Key words: multi-target tracking, probability hypothesis density filter, sequential Monte Carlo, track continuity, state extraction 