|Stochastic Systems Group|
Lei Chen - SSG, MIT
We propose techniques to extend our previous graphical model-based multisensor-multitarget data association approach into dynamic setting. One challenge to our previous approach brought by the dynamic setting is how to obtain the graph structure when the sensor-target coverage configuration is uncertain. For this incomplete organization issue, we propose a new approach to construct the graphical model based on the sensor coverage subregions, where each subregion is a disjoint surveillance area covered by a distinct set of sensors. Such a graph structure will ensure the correspondence constraints between measurements and targets and the correspondence constraints between targets and subregions. We demonstrate a toy example of multiple-target tracking in a (bearing-only) sensor network by combining our data association procedure with a particle filtering tracker. Then we turn to address the trade-offs between the association performance achieved and communication costs required on the sensor networks. We propose a communication-sensitive form of the parallel message-passing algorithm that is capable of achieving near-optimal performance using far less communication. We demonstrate the effectiveness of our approach with experiments on simulated data.
Problems with this site should be emailed to firstname.lastname@example.org