Cross or Wait? Predicting Pedestrian Interaction Outcomes at Unsignalized Crossings

TL;DR

This work tackles predicting pedestrian crossing outcomes at unsignalized crossings using a controlled distributed-simulator dataset that captures both objective factors (e.g., time to arrival $T_a$, waiting time $T_w$, zebra vs non-zebra locations) and subjective traits (SVO and AISS). It evaluates linear, SVM, RF, and MLP models to predict whether a pedestrian will cross ($PCross$) and, for crossing cases, their crossing initiation time ($CIT$) and crossing duration ($CD$), employing five-fold cross-validation and classifying metrics ($ACC$, $F1$) alongside regression metrics ($MAE$, $RMSE$). The neural-network-based MLP improves crossing-decision accuracy by 4.46% and F1 by 3.23% over the logistic-baseline, while RF and MLP achieve substantial reductions in $MAE$ and $RMSE$ for $CIT$ and $CD$ (up to ~31% and ~30% respectively). An ablation study demonstrates how feature availability (e.g., excluding $SVO$ or $AISS$) affects model performance, offering guidance for model selection under limited inputs and underscoring the importance of zebra presence, $T_a$, waiting time, and personality traits in prediction. These results have practical implications for enhancing AD systems with predictive cues for pedestrian interactions under varied feature availability.

Abstract

Predicting pedestrian behavior when interacting with vehicles is one of the most critical challenges in the field of automated driving. Pedestrian crossing behavior is influenced by various interaction factors, including time to arrival, pedestrian waiting time, the presence of zebra crossing, and the properties and personality traits of both pedestrians and drivers. However, these factors have not been fully explored for use in predicting interaction outcomes. In this paper, we use machine learning to predict pedestrian crossing behavior including pedestrian crossing decision, crossing initiation time (CIT), and crossing duration (CD) when interacting with vehicles at unsignalized crossings. Distributed simulator data are utilized for predicting and analyzing the interaction factors. Compared with the logistic regression baseline model, our proposed neural network model improves the prediction accuracy and F1 score by 4.46% and 3.23%, respectively. Our model also reduces the root mean squared error (RMSE) for CIT and CD by 21.56% and 30.14% compared with the linear regression model. Additionally, we have analyzed the importance of interaction factors, and present the results of models using fewer factors. This provides information for model selection in different scenarios with limited input features.

Figures (7)

• Figure 1: Illustration for the distributed simulator study (DSS). (a) A pedestrian from the driver’s view: the pink bubbles are the body markers representing the pedestrian. (b) An interaction example: a pedestrian is crossing the zebra and interacting with the vehicle to their right.
• Figure 2: Top view of the zebra (left) and non-zebra crossing (right) with the designated standpoints (blue $\times$ markers). The first marker shows the pedestrian’s standing point, and the second shows the curb of the virtual road. The grey rectangles are visual obstructions (bus stops).
• Figure 3: The (a) prediction accuracy and (b) F1 score versus time to arrival for logistic regression (LR), support-vector machine (SVM), random forest (RF), and multilayer perceptron (MLP) models.
• Figure 4: Box plots of crossing initiation time for groundtruth (GT), linear regression (LR), RF, and MLP models. Outliers are marked as diamond dots.
• Figure 5: Distribution of predicted crossing initiation time. The y-axis is the normalized density such that the total area of the histogram in each subplot is equal to 1.