An Adaptive FCW System Based on Driver Attention States

Driver distraction poses a significant safety risk, with evidence showing that lapses in attention are a major contributor to traffic accidents. Existing forward collision warning (FCW) systems fail to account for dynamic variations in driver attention and rely on a limited set of predefined operational modes. This study introduces an adaptive FCW system that adjusts its functionality in real-time based on the driver’s attention state. Using the YOLO v5 algorithm, the system identifies the driver’s eye position and head tilt. Based on the proposed quantitative calculations, this detection results are classified into three levels of attention: safe, distracted, and dangerous. The FCW system then dynamically modifies the warning threshold according to the attention state, enhancing safety and comfort. To improve computer vision recognition, a virtual dataset was developed to supplement existing datasets. Generated using Stable Diffusion, this dataset contains over 4500 virtual portraits with diverse attributes such as race, occupation, age, and gender, encompassing three eye positions (front, left, right) and two head orientations (left turn, right turn). Experimental results demonstrate that training the model with this augmented dataset significantly improves recognition accuracy by nearly 4%. Furthermore, a virtual driving experiment conducted on the BeamNG platform demonstrated that the proposed FCW system outperformed existing systems, with collision incidents reduced by 29.41% and user comfort improved by 58.50% compared to the previous systems. By integrating the driver’s attention state as a dynamic variable, this innovative system enhances the adaptability of FCW technology to human factors. It offers a human-centered perspective within the field of human factors engineering to advance intelligent driving systems and provides valuable insights for future research in this domain.