Pedestrian Trajectory Prediction for Autonomous Vehicle by Integrating IA-LSTM and MSFM

In this paper, pedestrian trajectory prediction is systematically explored by integrating the Intention-Attention-Long Short Term Memory Network (IA-LSTM) and Modified Social Force Model (MSFM). Firstly, a novel IA-LSTM is developed for pedestrian trajectory prediction, pedestrian intention (waiting/crossing), pedestrian-pedestrian interactions and pedestrian-vehicle interactions are considered. Secondly, a MSFM is proposed for pedestrian trajectory prediction, the influences of other pedestrians, vehicles and crosswalk boundary are taken into account. Finally, an integrated model based on the IA-LSTM and MSFM is developed. Moreover, traffic data is collected at an un- signalized crosswalk, and the parameters of the MSFM are calibrated by proposing the use of Maximum Likelihood Estimation (MLE). The experimental results indicate that the integrated model surpasses the existing methods, and the prediction accuracy is improved by more than 19%, which inspires confidence in the application of the integrated model in the autonomous vehicle field to enhance the safety of pedestrians.