A Driver Behavior Detection Method Based on Improved YOLOv11 and an Attention Mechanism
TL;DR
This study presents a methodology to identify driver’s distraction using a refined You Only Look Once (YOLO) model, denoted as YOLOv11, which demonstrates higher accuracy and robustness, making it suitable for real-world driver monitoring system (DMS) deployments.
Abstract
Driver’s distraction and fatigue are among the major contributing factors of traffic accidents. This study presents a methodology to identify driver’s distraction using a refined You Only Look Once (YOLO) model, denoted as YOLOv11.To address the inconsistent performance of earlier versions of YOLO, especially with regard to lack of systematic evaluations, this study proposes an improved YOLOv11 model. A mixed local channel attention (MLCA) module is further introduced to enhance small object feature extractions considering the use of Wise-Intersection over Union (IoU) v3 loss function to improve localization accuracy and training stability. Experiments demonstrated that this model outperforms competing models across all metrics, achieving 99.13% mAP at 0.5 and 82.54% mAP at 0.5:0.95, while also achieving minimal bounding box loss. The proposed model demonstrated higher accuracy and robustness, making it suitable for real-world driver monitoring system (DMS) deployments.