Abstract
Recently, various risk assessment frameworks have been applied to local roads used by pedestrians. In the present work, a collision probability estimation framework based on a pedestrian behavior generation method is proposed for various road environments. In the generation method, road environments are translated into spatial representations, and each pedestrian model generates probabilistic candidates for behavior which avoids specific regions, e.g., obstacles, road users, and roadways. Then, by evaluating the frequencies of these candidates, collision probability can be estimated. Applying a particle filter to the proposed framework, numerical experiments are conducted to confirm the suitability of the method.
Keywords: pedestrian behavior, collision probability, probabilistic model, particle filter