Advancing Driver Behavior Modeling and Improved Driving Safety in the Age of Mixed Vehicle Automation Levels
With the expectation of continuing to reduce the fatality rate and improve road safety, enormous efforts have been made by researchers and industries in the development of fully autonomous vehicles. However, not only because of the complex nature of driving, factors such as technical performance, cost barriers, public safety, insurance issues, legal implications, and government regulations also bring a large number of challenges in this process. Such factors clearly suggest it is more likely that early steps in this progression will be multi-functional vehicles with level 2 – partial automation or level 3 – conditional automation. It is expected that during this transition period, vehicles with different levels of automation and driver engagement will be mixed together to form large-scale traffic environments. Therefore, greater research and understanding are needed regarding the vehicle and driver monitoring in these mixed assistive driving scenarios to improve driving safety. Either by using the vehicle’s own sensors (e.g., Controller Area Network (CAN)-Bus, camera, accelerations, etc.) or combine information from other sources (e.g., Vehicle-to-Vehicle (V2V) communication, Vehicle-to-Everything (V2X) communication, etc.). Next-generation intelligent vehicles should have the ability to evaluate and understand the driver’s status, performance, and driving behavior. As a result, they could warn protentional risks, provide guidance when necessary (e.g., lane level guidance), and make essential adjustments or actions when critical. Three general research questions could be raised to achieve advancing driver behavior modeling and improved driving safety in the age of mixed vehicle automation levels, which are (i) how can we acquire sufficient data, (ii) how to evaluate and understand driving behavior, and (iii) how to deliver information to drivers. This dissertation presents the efforts focusing on the last two problems. A number of aspects regarding driving behavior modeling and understanding are discussed, which includes driving performance analysis using vehicle dynamic signals, exploring the effectiveness of transfer learning for risky lane change maneuver detection, a cause study of long-term lane change intention prediction, as well as estimate driver’s cognitive workload from the environment. After receiving prediction results, a driver visual guidance framework is designed to deliver warning or guidance information to drivers for better decision making. Taken collectively, these advancements contribute to improved driver analysis and modeling, environment understanding, improved safety, and therefore ensure a safe, comfortable, efficient driving environment during this transition period.