Modeling human road crossing decisions as reward maximization with visual perception limitations