PL-TARMI: A deep learning framework for pixel-level traffic crash risk map inference.

Area of Science:

• Urban planning
• Geographic Information Systems (GIS)
• Transportation engineering
• Artificial Intelligence (AI)

Background:

• Accurate citywide traffic crash risk mapping is crucial for preventing accidents.
• Fine-grained geographic traffic crash risk inference is challenging due to complex road networks, human behavior, and extensive data needs.

Purpose of the Study:

• To develop a deep-learning framework, PL-TARMI, for accurate, fine-grained traffic crash risk map inference.
• To leverage easily accessible data for cost-effective traffic safety analysis and prevention guidance.

Main Methods:

• Integration of satellite imagery and road network data.
• Combination with accessible data sources including Point of Interest (POI) distribution, human mobility, and traffic data.
• Development of a deep-learning model (PL-TARMI) for pixel-level risk map generation.

Main Results:

• PL-TARMI successfully generates pixel-level traffic crash risk maps.
• The framework effectively utilizes diverse, accessible data inputs.
• Experimental results on real-world datasets validate the model's effectiveness.

Conclusions:

• PL-TARMI offers a novel and effective approach to fine-grained traffic crash risk assessment.
• The framework provides more reasonable and cost-effective traffic crash prevention guidance.
• This method addresses the challenges of complex road networks and data requirements in risk mapping.