Edge-Computing Deployment and Evaluation of YOLOv10 for Driver Drowsiness Detection

Abstract

Driver drowsiness is one of the major contributors to road traffic accidents. Recent advances in computer vision have enabled the development of object detection models capable of identifying visual indicators of drowsiness, including prolonged eye closure and yawning. This study develops a driver drowsiness detection model using a pretrained YOLOv10 object detection framework and evaluates its performance when deployed on the NVIDIA Jetson Orin Nano edge computing platform. The evaluation focuses on both detection performance and computational resource utilization under thirteen simulated driving scenarios. Experimental results show that the proposed model achieved an average inference accuracy of 29.49%, with an average inference time of 46.31 ms. During inference, the average CPU utilization reached 69.02%, while GPU and memory utilization averaged 66.57% and 0.66 MB, respectively. The relatively low detection accuracy was primarily influenced by detection loss, light reflections on eyeglasses, insufficient illumination, and limited model generalization. Despite these limitations, the results demonstrate that the YOLOv10 object detection framework can be successfully deployed on resource constrained edge computing devices such as the NVIDIA Jetson Orin Nano. Future work should focus on improving the robustness of the detection algorithm, expanding dataset diversity, increasing model performance under low light conditions, and utilizing imaging devices with enhanced low light capabilities.