基于多信息的疲劳状态识别方法

A Multi-Information-Based Fatigue State Recognition Method

基于机器视觉的疲劳检测方法具有非侵入性、快速、准确、全天候可操作等特点,逐步成为国内外研究热点,但该方法容易受复杂光照、驾驶员位姿变化的影响。针对此问题,对复杂光照和位姿变化对驾驶员疲劳检测的影响进行了深入研究,提出基于实时增强约束局部模型的多信息疲劳检测方法。对采集得到的图像进行实时高动态范围增强处理;对增强后的图像进行驾驶员人脸建模,提取驾驶员的视线、眼部PERCLOS特征;最后建立基于贝叶斯置信网络的多信息融合的疲劳状态检测识别方法。实验结果表明,该方法对于复杂光照和位姿变化情况下的驾驶员疲劳状态检测具有较强的稳健性。

Human fatigue detection based on the machine vision methods is non-invasive, fast, and accurate and is unhindered by weather conditions. Owing to these advantages, this technique has gradually become a hot research topic worldwide. However, it is easily affected by complicated illumination and changes in the pilot position. To solve this problem, on the basis of previous studies on driver fatigue detection under complicated illumination conditions and postural changes, we propose a fatigue detection method based on the real-time enhanced constraint local model. First, the collected images are subjected to real-time high-dynamic-range enhancement. Then, the enhanced image is used to model the driver＇s face in order to extract his/her vision and percentage of eye closure characteristics. Finally, the fatigue state is detected and an identification method based on Bayesian confidence networks is established. Our experimental findings show that the proposed method robustly detects the fatigue states of drivers under complex illumination and change in position.