基于视觉特征的疲劳检测算法研究

发布时间：2019-01-20 11:48

【摘要】：近年来,伴随着汽车数量以及驾驶人数的增加,全世界面临着更为严峻的交通安全形势。数据表明,疲劳驾驶约占重大交通事故的10%-15%。疲劳检测对于降低交通事故发生率,保障交通出行安全,有着重要的意义和应用前景。如何获取驾驶员的疲劳特征,评估其疲劳程度,已成为国内外研究的热点。本论文研究了国内外疲劳检测算法,并针对现有算法的特征少,稳定性差及准确度低等弊端,提出了一种基于多视觉特征的疲劳检测算法。首先,为了提高疲劳检测的准确度和稳定度,在基于人脸检测及关键点定位的基础上,论文研究给出了多个视觉特征来进行疲劳检测,增加了特征向量的维度。这些特征包括眨眼频率、闭眼时长、打哈欠的频率、打哈欠的总时长、点头的频率、点头的总时长、摇头的频率以及摇头的总时长。然后,论文通过各机器学习模型,对采集到的特征样本数据集进行训练和交叉验证,并根据精确率、准确率、召回率等指标,优选出逻辑回归(Logistic Regression,LR)模型,提出了一种基于LR模型的多特征融合的疲劳检测算法。最后,论文搭建了实时的疲劳检测系统,以300帧为检测滑动窗口,实现每秒一次的疲劳检测效果。当系统检测出驾驶员处于疲劳状态时,将立即给予提醒和警告。同时,论文将本文给出的基于LR模型的多特征融合疲劳检测算法与现有的P80单特征检测及基于模糊系统多特征融合检测算法进行了性能比较。结果表明,无论是在检测的准确率,精确率还是召回率上,本文提出的基于LR模型的多特征融合疲劳检测算法都明显优于其它两种检测算法。
[Abstract]:In recent years, with the increase of the number of cars and drivers, the world is facing a more serious traffic safety situation. The data show that fatigue driving accounts for about 10-15 percent of major traffic accidents. Fatigue detection plays an important role in reducing the incidence of traffic accidents and ensuring the safety of traffic travel. How to obtain the fatigue characteristics of drivers and evaluate their fatigue degree has become a hot spot at home and abroad. In this paper, fatigue detection algorithms at home and abroad are studied, and a fatigue detection algorithm based on multi-vision features is proposed to overcome the shortcomings of the existing algorithms, such as few features, poor stability and low accuracy. First, in order to improve the accuracy and stability of fatigue detection, based on face detection and key point location, this paper presents several visual features for fatigue detection, which increases the dimension of feature vector. These features include blink frequency, eye closure time, yawning frequency, total yawning time, nod frequency, total nod duration, frequency of shaking head, and total duration of shaking head. Then, through the machine learning models, the paper trains and cross-validates the collected feature sample data sets, and selects the logical regression (Logistic Regression,LR) model according to the accuracy, recall rate and other indicators. A fatigue detection algorithm based on LR model and multi-feature fusion is proposed. Finally, a real-time fatigue detection system is built, which uses 300 frames as sliding window to realize fatigue detection once a second. Warning and warning will be given immediately when the system detects driver fatigue. At the same time, this paper compares the performance of the proposed multi-feature fusion fatigue detection algorithm based on LR model with the existing P80 single feature detection algorithm and fuzzy system multi-feature fusion detection algorithm. The results show that the proposed multi-feature fusion fatigue detection algorithm based on LR model is obviously superior to the other two detection algorithms in terms of detection accuracy and recall rate.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6;TP391.41

【相似文献】