Kernel Based Method for Distributed Feature Tracking of Real-world Targets

Increasingly, target tracking systems are comprised of distributed heterogeneous sensors as part of a system that is not necessarily optimized for a specific target. In order to increase the speed at which resources across these sensor networks can be allocated, data should be fused on-platform to the extent possible, without relying on a centralized data fusion center or high communication bandwidth. Complications in this process arise when the sensors have only partially overlapping fields of view, observe barriers that obscure the target, or record features that are highly dependent on other variables, such as the orientation of the target. In real-world scenarios, it may be unclear whether the data collected by a particular sensor will improve or degrade tracking results for the entire system. The Density Tracking Distributed Kernel Fusion (DTDKF) method ameliorates these difficulties within the constraints of distributed sensing while tracking the long-term and instantaneous distinguishing features of multiple targets. In this paper, we apply the DTDKF method to real-world experimental data from the ESCAPE II data collect and compare against a traditional multi-hypothesis tracker to further explore the advantages and limitations of this technique.