Algorithms for Asynchronous Track-to-Track Fusion

Algorithms for synchronous track-to-track fusion (T2TF) have been widely studied. For the optimal T2TF, it is critical to take into account the crosscovariances between tracks of the same target due to (i) the common process noise, and (ii) information feedback [2]. The optimal memoryless (without memory–“woM”) T2TF with no information feedback (T2TFwoMnf) was studied in [3], [11]. In [15], the complete set of information configurations and the optimal algorithms for the synchronous T2TF were presented. These are T2TF without memory (T2TFwoM) with no, partial and full information feedback (designated as T2TFwoMnf, T2TFwoMpf and T2TFwoMff, respectively), as well as T2TF with memory (T2TFwM) with no, partial and full information feedback (designated as T2TFwMnf, T2TFwMpf and T2TFwff). The information matrix fusion (IMF) [14, 8] is a special case of T2TFwM. The advantage of IMF over the optimal T2TF is that it does not require the crosscovariances between the local tracks, which greatly simplifies its implementation. However, IMF is optimal only when the fuser operates at full rate [8, 7]. For reduced rate, IMF is heuristic. As reported in [6], IMF has consistency problems for extremely large process noise levels; however for most tracking scenarios it is consistent and has good tracking accuracy. In the real world, synchronization cannot be usually achieved among distributed local trackers where local measurements are obtained and local tracks are updated at different time instants. In addition, the communication between local trackers and the fusion center (FC) is subject to possible delays, and thus the fusion of delayed tracks should also be addressed. In [5] the problem of the fusion of delayed tracks is converted and solved as the fusion of out-of-sequence measurements (OOSM). However, the algorithm deals only with the fusion of delayed synchronous tracks at full rate. A pseudo measurement approach of fusing asynchronous tracks can be found in [12]. In [13], three approximate algorithms for AT2TF were proposed. Later in the present research, they are evaluated by simulations and shown to have consistency problems. This is because the crosscorrelation between the central and local tracks due to information feedback is not accounted. In this paper, first, the optimal (under linear Gaussian–LG–assumption) synchronous T2TF algorithm is generalized for the asynchronous situation, where the information configuration of memoryless fusion with partial information feedback (feedback only to the central track) [15] is used. The resulting algorithm accounts exactly for the crosscovariances between the central and local tracks. It handles both the asynchronous sampling times of the local trackers and the fusion of delayed tracks, and guarantees the consistency of the fused estimates. The optimal algorithms for the more complicated AT2TF with memory is not considered here, due to the limited gain in tracking accuracy, especially when significant geometric diversity exists among the local