图像梯度下眼控机械臂的眼动追踪算法

Eye-tracking algorithm for eye-controlled manipulator under image gradient

针对能够通过人眼球运动控制其运动的眼控机械臂,提出了一种基于图像梯度的眼动追踪算法。首先通过60 Hz摄像头捕捉连续帧图像并采用n帧一次比较的方式代替逐帧比较方式。然后对捕捉到的图像进行灰度化处理以消除噪声点。接着利用Haar特征和Adaboost分类器将人脸区域提取出来。再根据三庭五眼的原则进行眼球区域粗略萃取,并进行部分区域单独放大灰度化处理。最后利用图像梯度,对目标函数设定阈值进行二值化,完成精确的眼球中心定位。实验表明,该方法的准确度和定位速度完全适用于眼控机械臂。

According to the eye-controlled robotic arm that can control the movement of human eyeballs,this paper proposes an eye tracking algorithm based on image gradient.Firstly,it captures continuous frames of images through a 60 Hz camera and makes the frame-time comparison method instead of the frame-by-frame method.Then it grays the captured image to eliminate noise points,and uses the Haar feature and Adaboost classifier to extract the face region.According to the principle of three chambers and five eyes,the eyeball area is roughly extracted,and part of the area is separately amplified and grayed.Finally,it uses the image gradient to binarize the threshold of the objective function and completes the precise eyeball center positioning.The experiment shows that the accuracy and positioning speed of the method are suitable for the eye-controlled manipulator.