计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (9): 199-206.DOI: 10.3778/j.issn.1002-8331.2001-0348

• 图形图像处理 • 上一篇    下一篇

考场环境下考生视线估计方法

柴旭,方明,付飞蚺,邵桢   

  1. 1.长春理工大学 计算机科学技术学院,长春 130022
    2.长春理工大学 人工智能学院,长春 130022
  • 出版日期:2021-05-01 发布日期:2021-04-29

Sight Estimation Algorithms for Examinee in Examination Room Environment

CHAI Xu, FANG Ming, FU Feiran, SHAO Zhen   

  1. 1.School of Computer Science Technology, Changchun University of Science and Technology, Changchun 130022, China
    2.School of Artificial Intelligence, Changchun University of Science and Technology, Changchun 130022, China
  • Online:2021-05-01 Published:2021-04-29

摘要:

考生异常行为的监测容易使监考人员产生视觉疲劳。借鉴监考人员发现异常的过程,提出一种可用于考场异常行为分析的视线估计模型。为了减少图像中视线的信息损失,采用注视向量表示视线的大小和方向。该模型分为生成器、视线合成模块、鉴别器,先将考生头部图像输入生成器生成注视向量,再将头部位置和注视位置输入到合成模块得到真实注视向量。将头部图像与上述所得的两种向量输入鉴别器中,其生成对抗模式达到最优时,可得到生成真实值的生成器模型。实验结果表明,在多个考场环境中,该方法的性能优于所对比的几种方法。其中与Lian等人方法相比AUC(Area Under Curve)提高了2.6%,Ang(Angular error)和Dist(Euclidean distance)分别有效降低了20.3%和8.0%。

关键词: 考场, 视线估计, 生成对抗网络(GAN), 注视向量

Abstract:

Monitoring of examinee’s abnormal behavior is easy to make the invigilator to feel visual fatigue. In this paper, model based on line of sight estimation is proposed to automatically detect the abnormal behavior in the examination room. In order to reduce the information loss of line of sight, gaze vector is used to represent the size and direction of line of sight. The model consists of three parts:gaze vector generator, gaze synthesis module and discriminator. Image of examinee’s head is given as input to the generator to generate the gaze vector, and then the head position and gaze position of the examinee are input into the synthesis module to obtain the real gaze vector. The head image and the two vectors obtained above are input into a discriminator, and when the generation adversarial mode is optimal, a generator model that generates real values can be obtained. The experimental results show that the performance of this method is better than the several methods compared in multiple test room environments. Compared with those of Lian et al, the results show that the Area Under Curve(AUC) index is increased by 2.6% while Angular error(Ang) and Euclidean distance(Dist) of the model are efficiently reduced by 20.3% and 8.0% respectively.

Key words: examination room, line of sight estimation, Generative Adversarial Networks(GAN), gaze vector