Learning the Pedestrian-Vehicle Interaction for Pedestrian Trajectory Prediction

TL;DR

This paper introduces a novel Pedestrian-Vehicle Interaction (PVI) extractor for neural networks to improve pedestrian trajectory prediction by effectively modeling pedestrian-vehicle interactions, validated on real-world urban traffic data.

Contribution

The paper proposes the PVI extractor, a new neural network component, applicable to both sequential and non-sequential models, enhancing pedestrian trajectory prediction accuracy.

Findings

PVI extractor reduces ADE and FDE in LSTM models by over 7% and 5%.

PVI extractor improves convolutional models' accuracy by over 2%.

Models with PVI better capture pedestrian-vehicle interactions.

Abstract

In this paper, we study the interaction between pedestrians and vehicles and propose a novel neural network structure called the Pedestrian-Vehicle Interaction (PVI) extractor for learning the pedestrian-vehicle interaction. We implement the proposed PVI extractor on both sequential approaches (long short-term memory (LSTM) models) and non-sequential approaches (convolutional models). We use the Waymo Open Dataset that contains real-world urban traffic scenes with both pedestrian and vehicle annotations. For the LSTM-based models, our proposed model is compared with Social-LSTM and Social-GAN, and using our proposed PVI extractor reduces the average displacement error (ADE) and the final displacement error (FDE) by 7.46% and 5.24%, respectively. For the convolutional-based models, our proposed model is compared with Social-STGCNN and Social-IWSTCNN, and using our proposed PVI extractor reduces the ADE and FDE by 2.10% and 1.27%, respectively. The results show that the pedestrian-vehicle interaction influences pedestrian behavior, and the models using the proposed PVI extractor can capture the interaction between pedestrians and vehicles, and thereby outperform the compared methods.