Learning-based preferential surveillance algorithm for persistent surveillance by unmanned aerial vehicles

In this paper, we present an algorithm called Learning-based Preferential Surveillance Algorithm (LPSA), which is specific to the problem of persistent surveillance by unmanned aerial vehicles. The novelty of the algorithm lies in providing the capability to a UAV to increase the probability of target detection through learning. The algorithm considers the neighborhood of a localized target as potential additional targets and motivates the UAV to verify if this is indeed true. Once a target is located in any of the neighborhood grid points, that point is visited more often than other points. Moreover, this technique uses the risk calculation method of an existing geometric reinforcement learning algorithm to reduce the frequency of visits to risky regions. Simulation results are used to demonstrate the effectiveness of the algorithm.