头戴式视线跟踪系统的一点标定方法  

ONE-POINT CALIBRATION METHOD FOR HEAD-MOUNTED EYE TRACKING SYSTEM

作　　者：侯树卫[1,2] 李斌[1] 夏小宝[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,安徽合肥230026 [2]中国华阴兵器试验中心,陕西华阴714200

出　　处：《计算机应用与软件》2014年第9期152-156,共5页Computer Applications and Software

基　　金：国家自然科学基金委-广东省联合基金重点项目(U0835002)

摘　　要：标定方法是视线跟踪技术中的关键环节,直接影响跟踪精度和用户体验。目前头戴式跟踪系统所使用标定方法,需要多个标定点进行标定。为更快、更方便地进行标定,提出一种方法,只需一个标定点,便可提取足够的标定信息完成标定过程。该方法可适用于目前的多种映射方法,如DLT方法、多项式方法、神经网络方法等,标定时间仅需10 s,精度可达1°,与多点标定相比,效率显著提高,而精度无明显差异。此外,使用一种新的神经网络模型:ELM(极端学习机)实现了神经网络标定方法,ELM的快速学习性能,显著缩短了神经网络训练时间。The calibration method affects the tracking accuracy and user experience directly, so it is a key link in gaze tracking technolo- gy. Current calibration method used by the head-mounted tracking system requires multiple calibration points to accomplish this process. In order to calibrate faster and more convenient, we present a method which only requires one calibration point for extracting sufficient calibration information to complete the calibration process. This method can be applied to a variety of mapping methods used at present, such as the DLT method, the polynomial method, and the neural network method, etc. The calibration time takes only 10 s, and the precision reaches 1 °. Compared with multi-point calibration, it significantly improves the efficiency with no noticeable difference in precision. In addition, we use a new neural network model, the ELM （extreme learning machine ）, to realise the neural network calibration. ELM＇ s fast learning performance remarkably shortens the training time of the neural network.

关 键 词：头戴式 视线跟踪 一点标定 神经网络 极端学习机 随机采样一致性 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]