Real-time and efficient eyes and mouth state detection: an artificial intelligence application based on embedded systems

摘要

It is reported that many traffic accidents are caused by fatigued driving. The state detection of eyes and mouth is usually employed to judge whether the driver is fatigue. However, the traditional image processing methods cannot achieve satisfactory detection accuracy due to the changes of illumination, head posture, and other factors in the actual environment. To date, although the methods based on deep learning have reached adequate accuracy in the tasks of object detection, they are obliged to rely on high hardware configuration to meet the real-time requirements. To achieve satisfactory detection accuracy in eyes and mouth detection task and obtain good real-time performance on embedded platforms such as NVIDIA Jetson TX2 by improving the object detection algorithm based on deep learning. In this paper, based on the original YOLOv3-Tiny structure, we not only added the structure of deep residual learning, but also calculated six new anchor boxes according to the training set using the K-means algorithm. We also used six data augmentation methods in the training set to improve the detection accuracy. The improved algorithm proposed in this paper can achieve satisfactory detection speed and accuracy on the TX2 embedded hardware configuration platform, validating that the ameliorated scheme is effective.