Stackelberg Surveillance

Bayesian Stackelberg game theory has recently been applied for security-resource allocation at ports and airports, transportation, shipping and infrastructure, modeled as security games. We model the interactions in a camera surveillance problem as a security game, and show that the Stackelberg equilibrium of this game can be formulated as the solution to a non-linear program (NLP). We provide two approximate solutions to this formulation: (a) a linear approximation based on an existing approach (called ASAP), and (b) a hill-climbing based policy search approximation. The first reduces the problem to a single (but difficult) linear program, while the second reduces it to a set of (easier) linear programs. We consider two variants of the problem: one where the camera is visible, and another where it is contained in a tinted enclosure. We show experimental results comparing our approaches to standard NLP solvers.