一种基于人脸序列模式的机车驾驶员疲劳检测方法

摘要

Fatigue is an important factor to cause traffic accidents. This paper surveys the existing fatigue detection methods based on computer vision, and proposes a new method for extracting fused Gabor features of human faces based on Gabor transformation. By combining it with the Hidden Markov Model, a new fatigue detection method for vehicle drivers based on the human face image sequences is proposed. Considering that the characteristics of human face appearances are different in fatigue and vigilance patterns, in our new algorithm, the Baum-Welch learning algorithm is used to learn the parameters of fatigue patterns of the Hidden Markov Modern first. And then, during the recognition stage, the face image sequence to be detected will be represented by the fused feature sequence through Gabor transformation, and thereafter, the Viterbi algorithm will be used to compute the probability of the fused feature sequence belonging to the fatigue patterns, which can be used for fatigue detection of the human face image sequence. Finally, simulations are done to test the performance of our newly proposed method by testing different human face image sequences, which consist of different postures. The results show that the proposed algorithm has better detection performance than that of the existing methods which are based on the single face image.%驾驶疲劳是导致事故的重要原因,本文在综述基于面部特征的机车驾驶员疲劳检测方法的基础上,提出了基于Gabor变换的人脸特征融合抽取模型,并在此基础上,结合隐马尔可夫模型(HMM)提出基于人脸图像序列的机车驾驶员疲劳检测方法.根据在疲劳和非疲劳状况下人脸模式特征的不同,首先利用Baum-Welch学习方法从疲劳图像序列训练学习得出疲劳模式下的HMM参数;然后,在疲劳模式识别时,把待识别的人脸图像序列表示成Gabor融合特征序列,再利用Viterbi算法计算该特征序列属于疲劳模式的概率值,从而实现对人脸图像序列的疲劳识别；最后,对各种姿态下的不同人脸图像序列数据进行了仿真测试.实验结果表明,与已有基于单幅人脸图像的疲劳识别方法相比,具有更好的疲劳识别性能.