Online webcam-based eye tracking in cognitive science: A first look
Online webcam-based eye tracking in cognitive science: A first look
复制标题
DOI：
10.3758/s13428-017-0913-7
复制
发表时间：
2018-04-01
影响因子：
5.4
通讯作者：
Weigelt, Sarah
中科院分区：
文献类型：
--
作者：
Semmelmann, Kilian;Weigelt, Sarah
在线实验正在认知心理学的许多领域兴起，作为经典实验室实验的可行替代或补充。虽然最近的研究涵盖了基于表现和反应时间的范式，但认知心理学的一种工具迄今为止尚未受到太多关注：眼球追踪。在这项研究中，我们使用了 Papoutsaki 等人最近推出的基于 JavaScript 的眼动追踪算法。 （国际人工智能联合会议，2016）与消费级网络摄像头一起研究在线眼动追踪从在线数据传导的共同优势中受益的潜力。我们将实验室内进行的三个任务（注视、追踪和自由观看）与在线获取的数据进行了比较，以分析前两个任务的空间精度以及第三个任务中众所周知的注视模式的可复制性。我们的结果表明，在注视任务中，实验室内数据表现出约 172 px（屏幕尺寸的 15%，3.94A 视角）的偏移，而在线数据的准确度稍低（屏幕尺寸的 18%，207 px），并且显示出较高的方差。在刺激运动期间具有恒定偏移的追踪任务也发现了相同的结果（实验室内 211 像素，在线 216 像素）。在自由观看任务中，与鼻子（9.71％）和嘴巴（4.00％）等其他关键区域相比，我们能够复制对眼睛（28.25％）的高注意力归因。总的来说，我们发现基于网络技术的眼动追踪适合所有这三项任务，并且相信所需的硬件和软件将不断改进，以实现所有认知心理学中更复杂的实验范式。
Online experimentation is emerging in many areas of cognitive psychology as a viable alternative or supplement to classical in-lab experimentation. While performance- and reaction-time-based paradigms are covered in recent studies, one instrument of cognitive psychology has not received much attention up to now: eye tracking. In this study, we used JavaScript-based eye tracking algorithms recently made available by Papoutsaki et al. (International Joint Conference on Artificial Intelligence, 2016) together with consumer-grade webcams to investigate the potential of online eye tracking to benefit from the common advantages of online data conduction. We compared three in-lab conducted tasks (fixation, pursuit, and free viewing) with online-acquired data to analyze the spatial precision in the first two, and replicability of well-known gazing patterns in the third task. Our results indicate that in-lab data exhibit an offset of about 172 px (15% of screen size, 3.94A degrees visual angle) in the fixation task, while online data is slightly less accurate (18% of screen size, 207 px), and shows higher variance. The same results were found for the pursuit task with a constant offset during the stimulus movement (211 px in-lab, 216 px online). In the free-viewing task, we were able to replicate the high attention attribution to eyes (28.25%) compared to other key regions like the nose (9.71%) and mouth (4.00%). Overall, we found web technology-based eye tracking to be suitable for all three tasks and are confident that the required hard- and software will be improved continuously for even more sophisticated experimental paradigms in all of cognitive psychology.