On multitrack fusion under unknown correlation

Abstract

The recent proliferation of distributed sensor networks makes the multitrack fusion under unknown correlation highly significant. There have been two major approaches developed to date for the multitrack fusion under unknown correlation: an approach based on maximum mutual information and another based on the minimum overestimate of covariance intersection. Unfortunately, the former is applicable only to two tracks while the latter becomes ineffective for multiple tracks with excessive overestimation. This paper presents solutions for fusing an arbitrary number of tracks under unknown correlation. First, a maximum bound of cross-covariance between two tracks of unknown correlation is computed analytically, such that the computed maximum bound is used for fusion with the fused track less conservative than the one provided by covariance intersection method. For more than two tracks, the above two-track method can be sequentially applied for fusion. However, the sequential fusion results in a sequence-dependent and suboptimal solution. Therefore, this paper presents several alternative solutions that provide a better trade-off between optimality and consistency in fusion. Simulation results are provided to demonstrate the effectiveness of the proposed method.