Stochastic surveillance strategies for spatial quickest detection

We present stochastic vehicle routing policies for detection of any number of anomalies in a set of regions of interest. The autonomous vehicle collects information from a set of regions and sends it to a fusion center. The vehicle follows a randomized region selection policy at each iteration. Using the collected information, the fusion center runs an ensemble of cumulative sum (CUSUM) algorithms in order to detect the presence of an anomaly in any region. We first determine optimal stationary policies that result in quickest detection of all anomalies. We then study an adaptive policy that assigns higher selection probability to a region with higher likelihood of an anomaly. We provide a comparative study of these policies.