Implementation of probabilistic risk estimation for VRU safety

This paper describes the design, implementation and results of a novel probabilistic collision warning system. To obtain reliable results for risk estimation, preprocessing sensor data is essential. The work described herein presents all the necessary preprocessing steps such as filtering, sensor fusion and target tracking. Risk estimation can be much more accurate if the behavior of road users is modeled correctly. Therefore, behavior models for individual road user (pedestrians, cyclists and drivers) are described in terms of probability density functions. These functions are based on statistical properties obtained from literature and experimental results. Preliminary results of road tests are presented in a scenario that includes a pedestrian and two cyclists. The results show that the approach successfully handles the unpredictable behavior of vulnerable road users and vehicles in both forward and sideward direction.