Pedestrian Path Prediction at Un-signalized Crosswalk by Integrating MSFM and Att-LSTM

In this paper, pedestrian path prediction at an un-signalized crosswalk is systematically investigated by integrating Modified Social Force Model (MSFM) and Attention Mechanism-Long Short Term Memory Network (Att-LSTM), and the possible conflicts between pedestrians and vehicles are focused on. Firstly, a novel MSFM for pedestrian path prediction is developed by taking the reaction to the crosswalk boundary, the evasion with the conflicting pedestrians and the collision avoidance with the vehicles into consideration. Secondly, an Att-LSTM is developed for pedestrian path prediction, and the interactions between the pedestrians and vehicles are considered for the first time. Finally, an integrating model for pedestrian path prediction based on the MSFM and Att-LSTM is developed. In addition, aerial video data of traffic states at an un-signalized crosswalk is recorded and analyzed. Maximum Likelihood Estimation (MLE) is proposed to calibrate the non-measurable parameters of the MSFM. The experimental results show that the integrating model outperforms the existing models and predicts the paths of pedestrians more accurately.