一种红外条件下的新型眼睛状态识别算法

A new eye-state recognition algorithm under infrared conditions

针对复杂光照、头部转动和戴眼镜对眼睛状态识别算法的影响,提出了一种基于红外图像和形态学的眼睛状态识别算法,通过带850nm窄带带通滤光片的红外摄像头采集眼睛图像,提取眼睛轮廓特征描述子和眼睛骨架图像的方向链码实现睁眼和闭眼判别,同时引入图像可信度判别机制,把眼睛划分为睁开、闭合和不可信3种状态,从而极大地降低了算法模型在恶劣环境中的虚警率。实验证明,本文算法模型对于复杂光照、头部转动和戴眼镜等均具有较高的鲁棒性,在可信眼睛图像集中,睁眼正确识别率达到了95.21%,闭眼正确识别率达到了92.03%,均高于其他几种常用的眼睛状态识别算法,同时每秒能处理200张以上眼睛图像,满足实际驾驶环境中实时性的要求。

To solve the reduction of eye state recognition accuracy caused by complex illumination, rotation of the head and wearing glasses, this paper presents an innovative eye state recognition algorithm based on infrared image and morphology. Firstly,an active infrared light source with wavelength of 850 nm and a narrow band-pass optical filter placed in front of the camera are used to collect eye images. Then, the feature descriptor of eye contour and the direction chain code of eye skeleton image are extracted to detect the driver eye state. The image credibility mechanism is introduced in this method. The eye state is divided into three categories：eye opening,eye closure and untrusted states,which greatly reduces the false alarm rate of the algorithm in harsh environment. The experiment results show that the algorithm has a high robustness to the complex illumination, the rotation of the head, wearing glasses, etc. In the credible eye images set,the recognition accuracies of eye opening and eye closure are up to 95.21 and 92.03 % ,respectively,which are higher than other commonly used methods. Meanwhile, it can cope with more than 200 eye images per second, so it meets the real-time requirements in the real driving environment.