Multi-target tracking and data association on road networks using unmanned aerial vehicles

A cooperative search and track algorithm for surveilling multiple road vehicles is presented for fixed-wing Unmanned Air Vehicles (UAVs) with a finite field of view. The road network is formed into a graph with nodes that indicate the target likelihood ratio (before detection) and position probability (after detection). Target measurement data is associated to either the likelihood ratio tracker or a Bayesian target tracker. Data association uses a similarity score generated by finding the earth mover's distance between the measurement and track probabilities. Two strategies for motion-planning of UAVs balance searching for new targets and tracking known targets. The first strategy is to loiter over the peak track probability to maximize information about a known target. The second strategy is to continue searching for new targets, returning to known targets only when the peak track probability becomes low. Results from numerical simulations are included to illustrate the performance of the algorithm and to quantify algorithm performance under the influence of added uncertainty in the detection and measurement of targets.