Driver drowsiness is the major cause of many traffic accidents. A study by the National Institute of Sleep and Vigilance showed that the majority of the accidents took place on fast tracks. 34% of these accidents are mainly due to lack of sleep. Indeed, the drowsiness of drivers appears mainly by a fall of vigilance with the appearance of various behaviors that represent clues for the decrease of reflexes like the presence of yawning, heaviness of the eyelids and the difficulty to keep the head in frontal position compared to the vision field. In this context, we are interested particularly in the proposal of new solutions allowing the automatic control of the driver drowsiness states. These solutions need to be non-invasive while being based only on the analysis of the visual indices. These indices are calculated starting from the spatiotemporal information of the contents of a video stream. The main goal of our work is to recognize driver abnormal behavior by analyzing the characteristics of the face. In fact, we have proposed a fusion system based on detection of yawn, detection of somnolence and the 3D head pose estimation. This fusion system is evaluated by the three databases and shows many success of our suggested approach.

驾驶员嗜睡是许多交通事故的主要原因。美国国家睡眠与警戒研究所的一项研究表明，大多数事故发生在快速轨道上。其中34%的事故主要是由于睡眠不足。事实上，驾驶员的困倦主要表现为警惕性下降，出现各种行为，这些行为代表了反射减弱的线索，例如打哈欠、眼睑沉重以及与驾驶者相比难以将头部保持在正面位置。视野。在这种情况下，我们对允许自动控制驾驶员困倦状态的新解决方案的提议特别感兴趣。这些解决方案需要是非侵入性的，同时仅基于对视觉指标的分析。这些索引是从视频流内容的时空信息开始计算的。我们工作的主要目标是通过分析面部特征来识别驾驶员的异常行为。事实上，我们已经提出了一种基于哈欠检测、嗜睡检测和 3D 头部姿态估计的融合系统。该融合系统由三个数据库进行评估，并显示了我们建议的方法的许多成功。