Development and Evaluation of Lane Hazard Prediction Application for Connected and Automated Vehicles (CAVs)

More than 37,000 fatalities occurred on U.S. roads in 2016. The number of both vehicle miles traveled and traffic accidents has increased in the past two years, and is mainly attributed to human error, mainly due to inattentiveness during hazardous driving situations. By crowdsourcing traffic data (e.g. vehicle position, speed, direction, etc.) through vehicle-to-vehicle (V2V) communication, connected vehicles (CV) can detect upcoming hazards at lane level with reasonable lead time. The work described in this paper aimed to develop and simulate an innovative V2V-based application to perform lane-level hazard prediction, and a corresponding driver response model. The concept of Lane Hazard Prediction (LHP) is to improve the mobility and safety of both individual users and the entire traffic system. LHP identifies the position of a downstream lane-level hazard (within seconds after it occurs) based on a spatial and temporal data mining and machine learning techniques. It then guides the LHP-equipped vehicles with recommended lateral maneuvers to avoid traffic jams resulting from the hazards. Simulation results demonstrate reliable hazard prediction, even when the V2V penetration rate is as low as 20%. A comprehensive evaluation of the developed LHP application from the perspectives of both user benefits and system benefits has been conducted over different CV penetration rates. The results demonstrate that the proposed LHP application can significantly improve both the safety and mobility performance of the equipped vehicles without compromising the mobility and safety performance of the overall traffic.