摘要
驾驶员疲劳驾驶是造成交通事故的主要原因之一,为解决该问题,提出一种新的基于机器视觉的驾驶员疲劳状态检测方法。根据驾驶员视频图像特点,采用基于肤色检测的Adaboost算法提取面部以及手部的感兴趣区域(Regionsofinterest,ROIs)。基于尺度不变特征变换(Scale invariant feature transform,SIFT)特征点匹配获取眼、嘴以及手部的SIFT特征点,据此得出面部以及手部特征参数。将Perclos、MClosed、Phdown以及SA 4个特征参数作为模型输入,疲劳度等级作为模型输出,建立三层BP神经网络模型,并应用贝叶斯正则化并结合动量梯度下降法较好地解决了传统BP人工神经网络训练高精度和预测低精度的过拟合现象。试验数据表明,该方法能够克服光照、背景、角度以及个体差异的影响,且疲劳检测的正确识别率达到99.64%。
Fatigue driving is one of the major causes of traffic accidents.In order to solve the problem,a new method based onmachine vision for driver fatigue detection is proposed.According to the characteristics of the driver’s video image,the regions ofinterest(ROIs)of face and hand are extracted by the Adaboost algorithm based on skin color detection.Based on SIFT(scaleinvariant feature transform)feature points matching,the SIFT feature points of the eye,mouth and hand are extracted,and the facialand hand feature parameters are obtained.The 4 characteristic parameters of Perclos,MClosed,Phdown and SA are used as modelinputs,and the fatigue grade is used as model output.Three layer BP neural network model is established.The Bayesianregularization and the momentum gradient descent method are used to solve the overfitting phenomena of the traditional BP neuralnetwork training with high accuracy and low prediction accuracy.The experimental data show that the method can overcome theinfluence of illumination,background,angle and individual difference,and the correct recognition rate of fatigue detection is 99.64%.
作者
刘明周
蒋倩男
扈静
LIU Mingzhou;JIANG Qiannan;HU Jing(School of Mechanical and Automotive Engineering, Hefei University of Technology, Hefei 230009)
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2019年第2期18-26,共9页
Journal of Mechanical Engineering
基金
国家自然科学基金资助项目(51375134)
