基于样本扩充和改进Lasso回归的视线估计

为了利用眼部特征进行准确的视线估计，提出了一种基于样本扩充和改进Lasso回归的方法，建立眼部特征与视线之间的映射关系.通过对小样本评分得到优质样本，进而完成样本扩充，利用改进的Lasso回归得到准确的视线估计模型.该方法对标定过程中的眨眼等干扰具有鲁棒性，受干扰后仍可保持相对较高的视线估计准确度.实验结果表明:标定过程无干扰，该方法视线估计准确度比传统方法提高11.25%；标定数据加入6.67%异常数据，该方法视线估计准确度比传统方法提高22.62%. 

Gaze Estimation Based on Sample Expansion and Improved Lasso Regression

In order to make use of eye features for accurate line-of-sight estimation, a method based on sample expansion and improved Lasso regression was proposed to establish the mapping relationship between eye features and line of sight. Quality samples were obtained by scoring all samples, and then sample expansion was completed. The improved Lasso regression was used to obtain an accurate line-of-sight estimation model. This method is robust for interference such as blinking in the calibration process, and can still maintain a relatively high accuracy of line-of-sight estimation with interference. The experimental results show that the accuracy of sight estimation of this method is 11.25% higher than that of the traditional method without interference; the accuracy of sight estimation of this method is 22.62% higher than that of the traditional method with 6.67% abnormal data in the calibration data.