驾驶员疲劳监测方法设计

Drowsy Driver Detection System

对各种疲劳驾驶监测方法进行了比较,指出了各种方法在驾驶员舒适性、监测结果准确性以及对环境背景抗干扰性方面存在的缺陷。提出基于PERCLOS法,利用红外光源、差分图像、神经网络辅助Kalman滤波器实现总体设计。采集图像使用红外光源,利用人眼对两种波长(850nm/950nm)红外光线的吸收率的明显差异,得到差分图像,同时避免了环境背景中其他波长光线的干扰,满足了全天候的要求;利用PERCLOS作为评价标准对驾驶员眼部的差分图像分析,实现了非接触的手段获取信息,解决了影响驾驶员舒适性的缺陷;利用神经网络辅助Kalman滤波器对监测对象进行跟踪,获得的有效差分图像达到17帧/s以上,保证了系统结果的实时性。试验表明,该系统对驾驶疲劳的检测准确率达到86.89%。

Fatigue is one of the most important reasons for traffic accidents. The drowsy driver detection system （DDDS）, steering attention monitor （SAM） etc. were compared. The deficiencies of drivers comfort, resultant veracity and anti-interfere were indicated. The scheme based on infrared, difference image, and Kalman filter was adopted. The pupils＇ absorptivity of infrared is different at the wavelengths of 850 nm and 950 nm. The interference Of other wavelengths light in the background was avoided, and difference images were obtained. With PERCLOS as evaluate standard and the untouched measure, the driver felt comfortable. Kalman tracker assisted by neural net was used to deal with the dynamics of head/eyes movements and made the scheme be real-time. More than 17 frames of effective difference images were obtained in a second. The results showed that the veracity of this scheme to detect driver drowsy is 86.89 %.