Probabilistic Interaction-Aware Occupancy Prediction for Vehicles in Arbitrary Road Scenes

摘要

This paper drafts a new interaction-aware approach to predict road user occupancy through vehicle-specific probabilities of presence. Instead of presenting another behavior prediction approach which tries to predict the most likely vehicle positions, the aim is to determine the likelihood of all feasible vehicle positions, called probability of presence. Thus the occupied area derived from this probability of presence blocks less space compared to existing approaches in the field of occupancy prediction. For this purpose, based on a physical model the future presence areas of a vehicle are fully calculated in a discrete consideration. Each movement possibility is evaluated taking into account the statistically typical driver behavior and interactions with static and dynamic objects. This gives rise to a qualified prediction of the future behavior of vehicles which blocks less space compared to all feasible future vehicle positions, like demonstrated in the numerical example.