For crowd cognitive-based routing behaviors, human vision is a key factor and should be properly evaluated. To model pedestrians’ vision, the paper proposes the mechanism of trap. A trap is merely a tuple of two elements and defines an operating scope and an operation. A trap’s operation will be executed for a person whenever the person is walking inside its scope and caught by it. Implementation issues are discussed and an efficient implementation is sought for the developed OpenCL-based simulation program. By letting traps’ operation to be “seeing certain rooms, doors, etc” and installing them at suitable places, human vision can be modeled conveniently and efficiently. To validate the proposed trap mechanism to some extent, the numerical experiments firstly investigate plausibility of individuals’ evacuation route. Then routing at the collective level is experimented by assuming pedestrians of different levels of map knowledge and vision.