精神疲劳实时监测中多面部特征时序分类模型

针对现有疲劳监测方法仅根据单帧图像嘴巴形态进行哈欠识别准确率低,采用阈值法分析眨眼参数适应性较差,无法对疲劳状态的过渡进行实时监测等问题,提出一种新的进行精神疲劳实时监测的多面部特征时序分类模型.首先,通过面部视觉特征提取张口度曲线与虹膜似圆比曲线;然后,采用滑动窗口分段、隐马尔可夫模型(HMM)建模等方法在张口度曲线的基础上构建哈欠特征时序并进行类别标记,在虹膜似圆比曲线的基础上构建眨眼持续时间时序并进行类别标记;最后,在HMM的基础上增加时间戳,以便自适应地选取时序初始时刻点并进行多个特征时序的同步与标记结果的融合.实验结果表明,本文模型可降低哈欠误判率,对不同年龄的人群眨眼具有很好的适应性,并可实现对精神疲劳过渡状态的实时监测.

Time-series classification model based on multiple facial feature for real-time mental fatigue monitoring

To solve the problems of low recognition accuracy in yawn detection based on single-frame image mouth morphology, poor adaptability in blink characteristics analysis through threshold method, and unable to real-time monitor the transition state of fatigue in computer vision based fatigue monitoring method, a new classification model based on multiple facial feature time sequences for real-time mental fatigue monitoring was proposed. First of all, generate mouth opening degree curve and iris circularity ratio curve through facial visual features extraction. Then, using sliding window method and HMM etc. to build yawn characteristics time sequences based on mouth opening curve and annotation them; build blink duration time sequences based on iris circularity ratio curve and annotation them. Finally, add a time stamp on HMM to adaptively select the initial time point of the next time sequences, to perform synchronization and classification result fusion of the two feature time sequences based on the time stamp. Experimental results show that the presented model can reduce yawn detection error rate, have good adaptability for blink characteristics of different age groups, and can monitor the transition state of mental fatigue in real-time.