Fatigue Driving Detection Method Based on Improved YOLOv8n

Aiming at the problems of too many model parameters and the imbalance between accuracy and real-time performance in the existing fatigue driving detection, a lightweight fatigue driving detection method based on SBG-YOLOv8n was proposed. Utilizing the lightweight convolutional neural network ShuffleNet-v2 as the backbone network for facial fatigue feature extraction, the number of parameters and complexity of the model are significantly reduced. Subsequently, the BiFPN structure is employed to optimize the feature fusion network, and GIoU is utilized to enhance the loss function, thereby improving the accuracy of localizing the fatigue feature area. Finally, PERCLOS and PMOT are adopted to distinguish the fatigue state of drivers. Experimental results demonstrate that compared with the base model YOLOv8n, the SBG-YOLOv8n model witnesses an 80% reduction in parameters and achieves a remarkable detection accuracy of 98.69%, thereby effectively balancing the trade - off between accuracy and real - time performance in fatigue driving detection.