多特征融合的驾驶员疲劳检测研究
Research on Driver's Fatigue Detection by Multi-feature Fusion

针对驾驶疲劳检测中面部特征定位及驾驶员疲劳状态判别方法判断存在的不足，提出了利用监督下降算法同时定位驾驶员的多个面部特征。在眨眼、哈欠及点头判断的基础上，提取驾驶员眨眼频率、哈欠频率及点头频率多个特征值建立疲劳检测样本数据库，并构建朴素贝叶斯分类器进行疲劳判断。当驾驶员出现疲劳驾驶时及时给以警告信息，以预防交通事故发生。在实际的驾驶环境视频测试结果中，驾驶员疲劳状态的判别平均准确率达到了94.87%，具有较好的性能。
Aiming at the deficiencies of facial features location and driver fatigue judgments in driving fatigue detection, a new method called supervised descend method was proposed to locate driver's face features simultaneously. Driver's eye blink frequency, yaw frequency and nodding frequency are extracted to build the fatigue detection sample database based on eye blink, yawn and nodding judgments, then a naive Bayesian classifier was constructed to judge the driver's fatigue state. If a driver appears fatigue state during driving, warning message would be given promptly to prevent traffic accidents. In the actual driving environment video test result, the average accuracy rate of the driver's fatigue detection achieved 94.87%, with good performances