Abstract:

A multi-modal trajectory prediction method of pedestrians at signalized intersections for autonomous vehicles was proposed to improve the driving safety of autonomous vehicles in the mixed traffic with pedestrian and vehicles. Firstly, considering the social attributes of the Social Generative Adversarial Network (SGAN) model, the pedestrian history trajectory was taken as the model input, the generator and discriminator were trained alternately, and the cross-entropy loss function was used to optimize the model, and then a pedestrian trajectory prediction model based on SGAN was proposed. Secondly, four binding force models based on pedestrian self-drive, pedestrian interaction, zebra crossing boundary force and traffic light force were established, and then a pedestrian trajectory prediction model based on Social Force Model (SFM) was proposed. The particle swarm optimization algorithm was used to calibrate the non-measurable parameters of SFM. Finally, based on the AdaBoost algorithm, the prediction results of SGAN and SFM were fused, and the weights of each model were iteratively trained and optimized dynamically by multiple weak learners to improve the prediction accuracy of the model. Based on the pedestrian data of an intersection in Xi'an city, the experimental analysis and verification were carried out. The results show that the average displacement error (ADE) and final displacement error (FDE) of the proposed method are increased by about 21.7% and 10.5%, respectively, compared with the single SFM model and the single SGAN model. The proposed model can realize more accurate pedestrian trajectory prediction.

Key words: autonomous driving, urban intersections, pedestrian, trajectory prediction, integration model