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

Abstract

A citywide traffic crash risk map is of great significance for preventing future traffic crashes. However, the fine-grained geographic traffic crash risk inference is still a challenging task, mainly due to the complex road network structure, human behavior and high data requirements. In this work, we propose a deep-learning framework PL-TARMI, which leverages easily accessible data to achieve accurate fine-grained traffic crash risk map inference. Specifically, we integrate the satellite image and road network image, combine with other accessible data (e.g., point of interest distribution, human mobility data, traffic data, etc.) as input, and finally obtain the pixel-level traffic crash risk map, which could provide more reasonable traffic crash prevention guidance with a lower cost. Extensive experiments on real-world datasets demonstrate the effectiveness of PL-TARMI.