Title Applying Artificial Intelligence Techniques to Identifying Accident-prone Road Sections (3/4): Analysis of Dangerous Events and High Risk Driving Behavior
Dept Transportation Safety Division
Year 2024
Month 7
Price 550
Summary        The executive summary presents an advanced project dedicated to enhancing road safety by broadening the scope of accident prevention strategies. Traditionally focused on proximal causes, this initiative advocates for a more inclusive analysis that encompasses not only accidents but also potential dangerous driving events and behaviors. Leveraging state-of-the-art in-vehicle and external image recognition technologies, this project—now in its third year of a four-year plan—builds upon prior research to pinpoint risk indicators for dangerous events and high-risk driving behaviors. By augmenting sample sizes, integrating cutting-edge technologies, refining image recognition methods, examining influential factors, and undertaking both broad and detailed spatiotemporal
analysis, the project aims to furnish educational training materials and safety management guidelines for operators and inform safety improvement strategies for road authorities.
       The project develops a comprehensive model to identify high-risk driving behaviors, categorizing management systems based on data completeness and incorporating various data sources, including vehicle system and global satellite positioning data, Advanced Driver Assistance Systems (ADAS), and external environmental data. A novel hybrid model combines quantitative economics with machine learning techniques to assess high-risk driving behaviors, establishing risk values and identifying hazardous scenarios.
Experimental results highlight the potential of image recognition technologies like YOLOv7 and YOLOv8, and ViTPose in accurately detecting internal and external risk factors, despite challenges such as overexposure and interference. Spatiotemporal analysis provides insights into warning event patterns across different management systems, revealing the impact of traffic volume, service areas, and scheduling on driver behavior.
       To address fleet management challenges, the project proposes an AI-powered system that streamlines data processing, risk evaluation, and reporting, promising significant efficiency gains for transportation operators. This system not only quantifies the risk of dangerous events but also offers detailed insights into driver behavior, enhancing overall fleet safety management.
Post date 2024/07/29
View count: 162