Sensor Data Fusion

Toyon proposes to develop a track filter that exploits stochastic simulation for accurately propagating a targets motion through time. Moreover, the filter will exploit classification knowledge to select the motion models used during stochastic simulation of the target dynamics. The stochastic simulation component will exploit a library of target-specific dynamic motion models that are customized for the variety of targets encountered. Each target class will have a defined dynamic motion model derived through knowledge of target motion models in addition to empirical analysis of collected data on these targets. The appropriate dynamic models are selected in real time based on the results of a target classification module, which exploits kinematic characteristics of the object in track, as well as contextual information, and exploits any a priori knowledge of the different targets operating parameters in order to classify the target (or at least restrict it to a subset of targets). Correlated random processes will be used to model any uncertainties in the dynamic model. The developed track filter will be evaluated on realistic sensor data on representative targets in order to demonstrate superior performance when compared with other track filters. Approved for Public Release | 16-MDA-8917 (15 November 16)