Road user profiling using multimodal data of naturalistic driving databases

◎Research at a glance

This research investigates how naturalistic driving data and artificial intelligence can be combined to detect safety- and risk-relevant driving patterns and develop comprehensive, personalized driver profiles. The research first addresses false tailgating detections caused by vehicles cutting into the driving lane by integrating object detection, object tracking, and lane detection using dashcam video. It then develops AI-based methods to assess driver behavior at both microscopic and macroscopic levels, including risky event detection, trip-level behavioral summarization, and the identification of longer-term driving habits and behavioral changes. Multimodal deep learning, CNNs, and large language models will be explored to transform multimodal driving data into interpretable driver profiles. Ultimately, the research aims to provide personalized safety recommendations and intelligent coaching-video recommendations based on previous algorithms.