基于人工神经网络动态标定算法的低成本视线追踪系统

针对视线追踪系统成本高、标定算法复杂的问题，研究了一种低成本视线追踪系统.系统采用低成本网络摄像头，采集到的图像首先采用Haar-like特征与肤色结合算法来进行人脸检测，并利用主动表现模型算法和光流法定位并跟踪人脸特征点；然后利用梯度向量法进行瞳孔中心检测；为了提高系统精度和鲁棒性，提出了一种人工神经网络的动态标定算法.实验表明，视线追踪系统不仅具有很好的鲁棒性，而且具有较高的精度，在头部静止的情况下平均误差为1.34°，在头部运动的情况下平均误差为3.26°. 

A Low-Cost Eye-Gaze Tracking System Based on Artificial Neural Network Dynamic Calibration Algorithm

In order to reduce the cost of eye-gaze tracking system and simplify the complexity of the calibration algorithm,a low-cost eye-gaze tracking system was developed.The Haar-like feature and skin color combination algorithm were used to detect the human face.The active appearance model (AAM) algorithm and the optical flow method were used to locate and track the face feature points.And the pupil center was detected by the gradient vector method.An artificial neural network dynamic calibration algorithm was proposed to improve the tracking accuracy and robustness.Experiments show that the eye-gaze tracking system not only has better robustness,but also has higher precision.The average error of the system is 1.34° at head rest,and 3.26° at head movement.