Tracking multiple closely spaced targets using an adaptive foveal sensor

We address the problem of configuring a foveal sensor to track multiple, closely spaced moving targets. The foveal sensor has a high acuity region, whose center and extent can be configured, surrounded by a low acuity region. We study three heuristic approaches to extend a near-optimal greedy configuration rule for a single target to multiple targets: simultaneously observe all targets (SO), center the foveal region on each target in turn (TO), and center the foveal region on the target with the worst position estimate (WO). The target tracker is implemented using a particle filter with joint probabilistic data association (JPDA). Additionally, we implement two different independent-partition proposal distributions using JPDA and global nearest neighbor (GNN). Monte Carlo simulations show that the WO rule outperforms the other rules and that the IP-JPDA proposal gives better tracking performance.