摘要
人脸定位是采用机器视觉非接触式方法监测列车司机驾驶状态的基础环节。针对疲劳驾驶监测对实时性和鲁棒性要求高的问题,提出一种融合检测与跟踪的人脸定位方法。在检测阶段,构造H-Cg Cr肤色模型并提出一种加速方法筛选人脸候检区域;构造融合LBP特征训练出具有旋转不变性的SVM分类器,检测被AdaBoost漏检或误检的人脸。在准确获取人脸位置后,启动fDSST算法进行跟踪,并对算法的跟踪置信度进行判定以确保跟踪效果。实验结果表明:该方法能提升检测率,降低误检率,当司机头部摆动、转动或发生部分遮挡时,其脸部均能被准确定位,鲁棒性强;肤色加速检测方法能将人脸候检区域筛选耗时降低81.09%,f DSST跟踪算法能将人脸定位平均速度提升至45帧/s,满足实时性要求。
Face location is the basis step of using non-contact method based on machine vision to monitor the driving state of train drivers. Aiming at the problem that fatigue driving Monitoring requires high real-time and robustness, A face location method based on fusion detection and tracking was proposed. In the detection stage, the H-CgCr skin color model was constructed and an accelerated method was proposed to screen face candidate regions. And a fusion LBP feature was constructed to train a rotation invariant SVM classifier to detect the face missed or mistakenly detected by AdaBoost classifier. After accurately obtaining the face position, the fDSST algorithm was started to track, and the tracking confidence of the algorithm was determined to ensure the tracking effect. Experimental results show that the proposed method can improve the detection rate, reduce the false detection rate, and accurately locate the driver’s face when his head swings, rotates or partially occludes, which is robust. The skin color acceleration detection method can reduce the screening time of face candidate regions by 81.09%. The fDSST tracking algorithm can improve the average speed of face location to 45 frames per second and meet the real-time requirements.
作者
陈忠
潘迪夫
韩锟
CHEN Zhong;PAN Difu;HAN Kun(School of Traffic and Transportation Engineering,Central South University,Changsha 410075,China)
出处
《铁道科学与工程学报》
CAS
CSCD
北大核心
2019年第12期3134-3142,共9页
Journal of Railway Science and Engineering
基金
湖南省自然科学基金资助项目(2016JJ4117)
