Driver Distraction Behavior Recognition for Autonomous Driving: Approaches, Datasets and Challenges

Driver distraction behavior recognition is currently a significant study area that involves analyzing and identifying various movements, actions, and patterns exhibited by drivers while operating vehicles. This field has received considerable attention due to its potential to enhance driving safety through driver monitoring tasks, widely implemented in advanced driver assistance systems and autonomous vehicles. As a result, extensive efforts have been made to utilize different sensor modalities and algorithms to understand and classify driver behavior. This paper provides a comprehensive overview of driver behavior recognition methods, with a particular focus on deep learning-based approaches, that encompass multiple data patterns. Both visual and non-visual methods are explored, and their advantages and disadvantages are compared and analyzed. Additionally, this paper investigates the current popular driver behavior datasets. It systematically categorizes these datasets based on visual cues, including the driver's hand, facial, and upper body movements, as well as non-visual modal signals. The performance of state-of-the-art driver behavior recognition approaches are presented and the corresponding challenges are also revealed. Finally, the future trends and directions in this field are summarized.