Target Tracking and Adversarial Reasoning for Unmanned Aerial Vehicles

Because of their ability to reach unique vantage points without endangering a human operator, camera-equipped unmanned aerial vehicles (UAVs) are effective tools for military and civilian surveillance missions, such as target tracking. However, visually tracking targets can be challenging because of the inherent clutter and occlusions. To add to this challenge, adversarial targets will attempt to escape. To counter these challenges a two tiered approach is used. In the first tier, a particle filter is used to estimate the location of the target using information from the incoming video stream. The particle filter is a sample-based tool for approximating the solution to the optimal, Bayesian tracking problem. The filter is adept at approximating non-Gaussian distributions that evolve according to non-linear dynamics. However, this increased functionality comes with an inherently large computational burden. A methodology for allowing the filter to manage the computational load of the filter based on the tracking conditions is presented along with simulation and flight test results. In the second tier, an adversarial reasoning module is used to produce strategies for a team of UAVs that is tracking an evading target. By using a differential game framework a team of air vehicles is able to contain a target that is attempting to escape. The framework decomposes a complete game into a set of two player games, which are solved more easily. The framework is presented along with simulation results.