Tracking in hyper-spectral data

This paper builds on previous presentations of the histogram probabilistic multi-hypothesis tracking (H-PM,HT) algorithm. The histogram model used in H-PMHT is extended to treat hyper-spectral data, i.e., data in which each spatial cell has spectral content. The general case in which each measurement scan consists of a multidimensional array of intensity values (e.g., the "data-cube" of hyper-spectral imaging systems) is treated. This data array is interpreted as a spatial-spectral histogram. Direct application of H-PMHT to such data would track local energy peaks in the spatial-spectral domain; however, this approach is sub-optimal when the spatial track is of principle interest and the energy source has significant spectral extent. Spectral H-PMHT assumes that the spectral characteristics of the sources are known and available in simple non-parametric forms. The additional structure of the spatial-spectral signal model is presented, followed by an outline of the Spectral H-PMHT algorithm. The improvement in spatial tracking due to the inclusion of spectral information is demonstrated using simulated intensity data on bearing-frequency cells.