A Novel Prediction Algorithm of Pedestrian Activity Region for Intelligent Vehicle Collision Avoidance System

With the development of intelligent vehicle technology, pedestrian trajectory prediction techniques play a key role in collision avoidance systems. Activity regions are commonly adopted to cover the uncertainty of predicted trajectories. However, setting a rational region size for each pedestrian is a challenging issue. Excessive region size may lead to over-reactions of vehicles and limit the degree of freedom of the path planning while insufficient region size may fail to cover the uncertainty. This work proposes a novel dynamic activity region (DAR) prediction algorithm and a relevant collision risk evaluation method. In the prediction module, the proposed method predicts a rational activity region for each pedestrian based on the historical trajectory in contrast to the conventional methods which adopt preset static activity regions (SAR) for all pedestrians. The comparative study shows that the proposed method reduces the sizes of pedestrian activity regions by 10.5% on average with similar precision. In the cases where the trajectory is highly nonlinear, the region size reduction can reach 12.1%. In the collision risk evaluation module, the risk statuses of pedestrians are classified based on the relationship between their DARs and the future trajectory of the vehicle. This method can produce a more detailed classification. The effectiveness and improvement of the proposed method are validated by experiments and simulations.