Pedestrian Crossing Intention Forecasting at Unsignalized Intersections Using Naturalistic Trajectories.

Area of Science:

• Computer Vision and Machine Learning
• Robotics and Autonomous Systems
• Transportation Engineering

Background:

• Autonomous vehicles face challenges interacting with road users, especially pedestrians in urban environments.
• Current systems are reactive, responding only when a pedestrian is already present, leading to safety concerns.
• Anticipating pedestrian crossing intent is vital for proactive safety measures and smoother autonomous vehicle operation.

Purpose of the Study:

• To formulate pedestrian crossing intent forecasting at intersections as a classification problem.
• To propose a novel model for predicting pedestrian crossing behavior around urban intersections.
• To provide a quantitative confidence level (probability) alongside the predicted crossing intention.

Main Methods:

• Developed a classification model to predict pedestrian crossing behavior.
• Utilized naturalistic trajectory data from a publicly available drone-recorded dataset for training and evaluation.
• Evaluated the model's performance in forecasting crossing intentions at various locations within an urban intersection.

Main Results:

• The proposed model successfully predicts pedestrian crossing intention.
• The model provides both a classification label (crossing/not-crossing) and a probability score.
• Accurate predictions were achieved within a 3-second time window.

Conclusions:

• Anticipating pedestrian intent significantly enhances the safety and efficiency of autonomous vehicles.
• The developed model offers a reliable method for proactive pedestrian behavior prediction.
• This approach represents a significant step towards safer urban autonomous navigation.