摘要
现实中基于图像处理的疲劳驾驶监测往往因环境的变化而具有不确定性。监测算法不规范,以致于疲劳驾驶监测任务很具有挑战性。为了解决此问题,提出了一种基于多算法融合的动态滑动窗口算法框架。首先利用Adaboost算法识别人眼,然后改进Otsu算法来自适应各种不同环境;进而提出动态滑动窗口算法来得到睁闭眼之间的最佳阈值;最终,利用改进的PERCLOS算法估计疲劳驾驶状态的不同级别。针对环境的变化采用睁闭眼判断窗口随人眼特征变化而更新的策略,系统使用摄像头实时捕获人眼图像,并在PC机上进行仿真测试,可在130-150ms之间实现不同疲劳状态的识别。实验结果表明,此算法框架能够有效、快速的分辨驾驶员不同的疲劳状态。
Fatigue driving detection based on image processing in practice often shows uncertainly be- cause of the ehauge in environments. The lack of mormalization in monitoring algorithm makes the fatigue driving detection task very challenging. In order to solve this problem, this paper proposed a dynamic slid ing window algorithm framework based on muhi-algorithm fusion. This algorithm framework firstly recog nizes human eyes with AdaBoost algorithm, then an improved Otsu algorithm is modified to automatically a dapt to varied environments. Furthermore, it proposes an effective algorithm based on dynamic sliding win dow in order to compute optimal threshold between open and close eye window. Finally, it cstimates the different level of fatigue driving with improved percentage of eyelid closure time (PERCLOS) algorithm. According to changes in the environment, the algorithm framework adopts the strategy that eyes judgment window updates following eyes features variation,human eyes images are captured with camera in real time,and the proposed method is simulated on personal computer to recognize different level of fatigue driving be- tween 130 ms to 150 ms. This paper presents a new learning strategy and multi-algorithm framework. The comparative experiments demonstrate that the proposed algorithm framework can effectively discriminate different level of fatigue state in driving by eye state tracking.
出处
《太原理工大学学报》
CAS
北大核心
2016年第4期518-522,共5页
Journal of Taiyuan University of Technology
基金
国家自然科学青年基金资助项目(61402318)
2015年广东高校省级重点平台和重大科研项目(2015KQNCX211)
2015年度广东省前沿与关键技术创新专项资金(重大科技专项)项目(2015B010108003)
2014年北京理工大学珠海学院校级科研发展基金项目(XK-2014-02)