Surveillance Evasion Through Bayesian Reinforcement Learning