Autonomous vehicles aim to improve driving safety and comfort. In autonomous car SAE level-3 operations, it is necessary to determine whether the driving authority can be transferred from the computer to the driver. The driver must be awake and sufficiently alert to switch to manual driving operation. Physiological measurement methods require sensors that are in contact with the human body. These sensors are annoying, frustrating, and inconvenient for the driver. The purpose of this study is to determine a driver’s condition using eye-closing time and yawning frequency by a visible-light camera and a thermal camera. Eye-closings and yawns were detected using an appropriate non-contact detector with a visible camera and thermal camera. When a visible-light camera was used, the driver state recognition rate was 90 % or higher when the driver’s surroundings were brightly lit, but the recognition rate decreased significantly to approximately 4 % when the surroundings were dark. Using a thermal camera, the face recognition rate was 74 % under bright and dark conditions. For the thermal images, DLib and OpenCV could be performed well. Therefore, the DLib and thermal image combination could be used for a reliable driver drowsiness detection task.

自动驾驶汽车旨在提高驾驶安全性和舒适性。在自动驾驶汽车 SAE 3 级操作中，需要确定驾驶权限是否可以从计算机转移到驾驶员。驾驶员必须清醒并且足够警觉才能切换到手动驾驶操作。生理测量方法需要与人体接触的传感器。这些传感器对驾驶员来说是烦人的、令人沮丧的并且不方便的。本研究的目的是通过可见光相机和热像仪使用闭眼时间和打哈欠频率来确定驾驶员的状况。闭眼和打哈欠使用合适的非接触式探测器和可见光摄像机和热像仪进行检测。当使用可见光相机时，当驾驶员周围环境明亮时，驾驶员状态识别率为 90% 或更高，但当周围环境较暗时，识别率显着下降至约 4%。使用热像仪，明暗条件下人脸识别率为74%。对于热图像，DLib 和 OpenCV 可以很好地执行。因此，DLib 和热图像组合可用于可靠的驾驶员睡意检测任务。