红外图像的疲劳状态识别方法

Detection Method for Fatigue State Recognition of Infrared Images

为了解决光照变化对疲劳检测系统造成的识别准确性不高的问题,提出了一种近红外环境下判断人眼状态的方法,即针对红外光补图像的人眼状态判断;首先,利用Adaboost算法进行人眼区域定位,在网格法标记人眼瞳孔部分的基础上,采用Retinex算法对红外图像进行增强;接着,结合"亮瞳效应"特性,对二值化以及边缘检测后的红外图像分别进行网格法闭合度计算,得到人眼闭合度大小;最后,根据闭合度计算结果设定双阈值并结合PERCLOS准则来判断人眼特征状态;另外,在DM642硬件平台上进行疲劳检测试验,实验结果表明,该方法的人眼状态识别率达到了90%以上,且平均每秒能处理21帧图片;证明了该方法不仅能有效解决光照变化带来的问题,而且满足疲劳状态检测系统的快速性、准确性和有效性等要求。

In order to solve the problems of low accuracy caused by light changes in fatigue detection system, a method is proposed to judge the state of eyes in near infrared environment, i. e. , it is a detection method for eyes state recognition of infrared images. First of all, it uses the grid method to mark eye＇s pupil and uses Retinex algorithm to enhance the infrared image based on the human eyes region located by Adaboost algorithm. Then, a grid method is adopted to calculate the closure of eyes after binaryzation and edge detection respective, which is related to bright pupil effect. Finally, the state of eyes is determined by setting the dual-threshold based on the results of the clo sure of eyes, which is combined with PERCLOS. Besides, The tests on the hardware platform of DM642 shows that the human eyes recognition rate is more than 90%, and the average processing speed is 21 images per second. It has proved that the method can not only solve the problems caused by light changes, but also meet the requirement of rapidity, accuracy and validity of the fatigue detection system.