智能眼镜人机交互功能中眼行为识别方法的设计与实现  

Eye movement detection of human-machine interaction for smart glass

作　　者：周巡 黄保青 陈亚晨 张敏娟[1,2] 康宏向 ZHOU Xun;HUANG Baoqing;CHEN Yachen;ZHANG Minjuan;KANG Hongxiang(Information and Communication Engineering College,North University of China,Taiyuan 030051,China;Institute of Radiation Medicine,Academy of Military Medical Science,Taiyuan 030051,China;Academy of Military Science PLA China,Bejjing 100850,China)

机构地区：[1]中北大学信息与通信工程学院,太原030051 [2]中北大学山西省光电信息与仪器工程技术研究中心,太原030051 [3]中国人民解放军军事科学院军事医学研究院辐射医学研究所,北京100850

出　　处：《激光杂志》2022年第3期76-81,共6页Laser Journal

基　　金：国家自然科学基金资助项目(No.61975188);国防科技创新特区项目。

摘　　要：针对目前智能眼镜产品人机交互功能中存在的响应速度慢、交互效率低、鲁棒性不好等问题,设计并实现一种眼行为识别方法。该方法利用智能眼镜中眼部摄像头能够稳定拍摄人眼图像的特点,使用模板匹配定位眼睛在图像中的位置,对剪裁后的眼睛图像使用矩阵遍历法和灰度积分投影,再结合人眼的行为规律判定凝视、眯眼、闭眼、双眨眼、上扫视、下扫视、右扫视、左扫视八种眼行为。实验结果表明,移植在树莓派上的眼行为识别系统对不同眼行为平均识别准确率为96.25%,平均处理速度为每帧11.25 ms,有较高准确率和实时性,与目前主流智能眼镜的人机交互方式相比,具有低成本、易搭建和不受姿态影响等特点。In order to solve the complex human-machine interaction problems in current smart glasses products,this paper proposed an eye movement detection method and built an eye movement detection system.Since the eye camera in smart glasses can capture human eye images stably,the method used template matching to locate the position of the eyes in the image first,and then processed the cropped eye image based on matrix traversal and integral projection.Finally,based on the behavioral characteristics of the human eye,the system achieved eight types of recognition of eye movement,such as Gaze,Squint,Close,Double blink,Glance up,Glance down,Glance right and Glance left.Experimental results show that the system has high accuracy and real-time.The average correct rate of this eye movement detection system transplanted on the Raspberry Pi is as high as 96.25%,and the average detection speed is 11.25 ms per frame.Compared with the current mainstream smart glasses and eye-machine interaction products,the system has the characteristics of low cost,easy to build and not affected by posture.

关 键 词：眼行为识别 眼机交互 眼控装置 眼行为 人机交互 智能眼镜 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]