Predicting Pedestrian Crossing Behavior in Germany and Japan: Insights into Model Transferability

TL;DR

This study addresses cross-country differences in pedestrian crossing behavior and model transferability between Germany and Japan using VR-sim data. It evaluates four prediction approaches—Linear, SVM, RF, and NN—for gap selection, zebra crossing usage, and trajectories, and introduces unsupervised clustering to boost cross-country transfer. Key findings show that neural networks provide superior performance and transferability for gap and zebra-usage tasks, while random forests excel in trajectory prediction; Japanese pedestrians exhibit more cautious behavior, often selecting larger gaps and waiting longer. The work demonstrates that clustering-based training and country information as features can yield transferable, interpretable predictors suitable for intelligent-vehicle systems across different national contexts.

Abstract

Predicting pedestrian crossing behavior is important for intelligent traffic systems to avoid pedestrian-vehicle collisions. Most existing pedestrian crossing behavior models are trained and evaluated on datasets collected from a single country, overlooking differences between countries. To address this gap, we compared pedestrian road-crossing behavior at unsignalized crossings in Germany and Japan. We presented four types of machine learning models to predict gap selection behavior, zebra crossing usage, and their trajectories using simulator data collected from both countries. When comparing the differences between countries, pedestrians from the study conducted in Japan are more cautious, selecting larger gaps compared to those in Germany. We evaluate and analyze model transferability. Our results show that neural networks outperform other machine learning models in predicting gap selection and zebra crossing usage, while random forest models perform best on trajectory prediction tasks, demonstrating strong performance and transferability. We develop a transferable model using an unsupervised clustering method, which improves prediction accuracy for gap selection and trajectory prediction. These findings provide a deeper understanding of pedestrian crossing behaviors in different countries and offer valuable insights into model transferability.

Figures (12)

• Figure 1: Schematic overview of the experimental environment. Start (yellow drop) and goal (green square) are visualized and alternated between north and south sides of the road in every other trial. Cars were approaching from both directions with randomly selected gaps per lane.
• Figure 2: The boxplot of the predicted accepted gaps for different models. GT for ground truth, LR for linear regression, RF for random forest, and NN for neural network.
• Figure 3: (a) Distributions of the number of unused car gaps for both lanes. The pedestrians from the study conducted in Japan missed more car gaps than those from Germany. (b) The mean value of the accepted gap versus the number of unused car gaps for both lanes. DE stands for data collected from Germany, and JP stands for data collected from Japan. For data from both countries, as the number of unused car gaps increases, pedestrians tend to accept smaller gaps.
• Figure 4: The mean value of the accepted gap versus the largest missed car gap for both lanes. For data from both countries, as the largest missed car gap increases, pedestrians tend to accept smaller gaps.
• Figure 5: (a) Distributions of the pedestrian waiting time. Pedestrians from the data collected in Japan tend to wait longer than participants in the study conducted in Germany. (b) The mean value of the accepted gap versus pedestrian waiting time. For data from both countries, pedestrians tend to select larger gaps when waiting longer.