Pedestrian Behavior Prediction based on Motion Patterns for Vehicle-to-Pedestrian Collision Avoidance

This paper proposes a prediction method for vehicle-to-pedestrian collision avoidance, which learns and then predicts pedestrian behaviors as their motion instances are being observed. During learning, known trajectories are clustered to form Motion Patterns (MP), which become knowledge a priori to a multi-level prediction model that predicts long-term or short-term pedestrian behaviors. Simulation results show that it works well in a complex structured environment and the prediction is consistent with actual behaviors.