Recursive filtering for target tracking in multi-frame track-before-detect

This paper addresses the tracking problem with respect to dynamic programming (DP) based track-before-detect (DP-TBD). Basically, DP-TBD is a grid-based method performing search in the discretized state space, therefore the accuracy of its estimation suffers at least half a grid loss. Moreover, different from the conventional detection procedure which provides isolate point plots to a tracker, DP-TBD is a batch processor whose detection outputs are short state sequences in time series. Essentially, these short tracks are neither filtered target trajectories nor conventional point detections. Meantime the false alarms and missing reports of these short tracks also exist especially when target signal-to-noise ratio (SNR) is low. Thus, a tracker which can associate, smooth and finally combine these short tracks into continuous target trajectories is needed. In this work, we propose a two-stage, i.e., prediction and updating stages, recursive filtering algorithm for DP-TBD. The key idea is using the outputted short tracks after every DP-TBD batch processing as the input measurements for the updating stage of filtering. Simulation results show that the proposed algorithm can correctly estimate target trajectories and significantly enhance the tracking accuracy compared with the method without our filtering algorithm.