ABSTRACT
Pedestrian trajectory prediction plays an important role in easing traffic congestion and optimizing intelligent vehicle driving decisions. Aiming at the problem of pedestrian trajectory prediction, this paper establishes a deep learning model that can combine pedestrian collision avoidance information based on the interaction between pedestrian movements. The Voronoi diagram method and velocity obstacle method are used to divide the geometric space and velocity field of pedestrians into personal walking space. The interaction between pedestrians and other subjects is reflected by the change in pedestrian walking space. According to the two collision avoidance mechanisms, it is judged whether the pedestrian's motion track will enter into the area where a collision may occur in the future. Based on this, a long-short-term memory network is established. The acceleration of pedestrian historical moments and personal walking space data are used as inputs in an interaction module to describe the impact of pedestrian interaction on the selection of velocity. The information output by the interaction module and the movement state of the pedestrian are used together as the input of the long-short-term memory network to predict the future trajectory of the pedestrian. Finally, the prediction accuracy of the network is verified by experiments datasets studying the dynamics of pedestrians. The results show that the pedestrian walking space data can effectively improve the accuracy of pedestrian trajectory prediction. This also confirms that it is effective, to some extent, to reflect the impact of pedestrian interaction by calculating the change in pedestrian walking space.
