基于支持向量机回归模型的视线定位算法研究.pdfVIP

学兔兔 第35卷 第l0期 仪 器 仪 表 学 报 Vol_35 No．10 2014年 1O月 Chinese Journal of Scientific Instrument 0ct．2014 基于支持向量机回归模型的视线定位算法研究术 闰 蓓，王晓晶，蒋兆国 (北京航空航天大学自动化科学与电气工程学院 北京 100191) 摘 要：针对传统的基于单瞳孔一普尔钦光斑向量和多项式拟合的视线定位方法头动受限的问题，提出一种新的采用单摄像机 多辅助光源的基于二维映射模型的视线定位方法。利用4个瞳孔．普尔钦光斑向量构造系统输入特征向量，基于支持向量机回 归模型建立注视点映射模型定位视线，解决了视线定位系统的头部运动和系统非线性映射误差问题。实验结果表明：该方法与 传统方法相比，可以有效解决传统方法头动受限问题，适应性更强且精度更高，平均定位精度为1．2。 关键词：视线定位；支持向量回归；瞳孔一角膜反射光斑向量 中图分类号：TH89 TP391．41 文献标识码：A 国家标准学科分类代码：520．2040 Study on the gaze estimation algorithm based on support vector regression model Yan Bei，Wang Xiaojing，Jiang Zhaoguo (School of Automation Science and Electrical Engineering，Beihang University，Beijing 100191，China) Abstract：This paper puts forward a novel two—dimensional mapping model based gaze estimation algorithm，which adopts single camera and multiple auxiliary light sources to solve the constrained head problem in the traditional gaze estimation algorithm based on single pu— pil center cornea reflection vector and polynomial fitting method．The method uses four pupil center cornea reflection(PCCR)vectors to construct the input characteristic vectors，and based on the support vector regression(SVR)model the gaze estimation mapping model is built to calculate the user’S line of sight．This method solves the head movement problem and reduces nonlinear mapping error．Experi— ment results show that compared with traditional algorithm，this method has higher positioning precision and stronger adaptability，and the average positioning accuracy is 1．2。． Keywords：gaze estimation；SVR；PCCR vector