STRIDE: Street View-based Environmental Feature Detection and Pedestrian Collision Prediction

Area of Science:

• Computer Vision
• Autonomous Driving Systems
• Urban Planning

Background:

• Autonomous driving systems require enhanced environmental awareness to prevent pedestrian injuries.
• Predicting pedestrian collisions is crucial for developing safer autonomous vehicles.
• The impact of the built environment on pedestrian safety is not fully understood.

Purpose of the Study:

• To introduce a novel benchmark for studying the relationship between built environment elements and pedestrian collision prediction.
• To enhance environmental awareness in autonomous driving systems for active injury prevention.
• To establish a foundation for understanding built environment influences on pedestrian safety.

Main Methods:

• Introduction of a built environment detection task using large-scale panoramic images.
• Development of a detection-based pedestrian collision frequency prediction task.
• Proposal of a baseline method integrating a collision prediction module into a state-of-the-art detection model for simultaneous task tackling.

Main Results:

• Demonstrated a significant correlation between the detection of built environment elements and pedestrian collision frequency prediction.
• Established a baseline method capable of addressing both detection and prediction tasks concurrently.
• Provided empirical evidence for the interdependencies between built environment conditions and pedestrian safety.

Conclusions:

• The developed benchmark and methods offer a stepping stone towards improved pedestrian safety in autonomous driving.
• Object detection of built environment elements is a key factor in predicting pedestrian collision frequency.
• Further research into these interdependencies can lead to more robust and safer autonomous driving systems.