摘要
眼部状态的检测被认为是目前最准确可靠的疲劳驾驶状态识别方法,快速、准确地定位眼部是疲劳检测的关键。针对AdaBoost算法的训练过程中的样本权重过拟合问题进行了改进,提出了新的权重更新方法,即每一轮设定一个权重更新的阈值,避免了样本权重过拟合现象的出现。并针对影响检测速度的要素,结合肤色模型,缩小了搜索范围,减小了检测的错误接受率,并提高了检测速度。在Linux系统下采用QT编程实现了眼部疲劳检测的整个过程。实验结果验证了该算法定位和检测的准确性、实时性。
The detection of eyes state is regarded as the most accurate and reliable method in identifying the state of fatigued driving,the key of fatigue detection is to locate eye rapidly and accurately.According to over fitting problem of the sample weight which occurred in the training process of AdaBoost algorithm,a new weight updating method was proposed to improve it,which was to set up a weigh updating threshold in each round,thus the phenomenon of sample weight over fitting was avoided.Based on the complexion model,especially for the factors affecting the speed of detection,the new method narrowed the scope of the search and reduced the false tolerance rate,and boosted the detection speed.The whole process of eye fatigue detection was realised with QT programming in Linux system.Experimental results proved the accuracy and real time of localization and detection of this algorithm.
出处
《计算机应用与软件》
CSCD
2011年第3期106-109,118,共5页
Computer Applications and Software
基金
浙江省科技厅面上项目基金(2007C30008)
