人類在認知學習的過程中，大部分的訊息是透過眼睛視覺所獲得，並且在視線範圍內若能找到感興趣之區域，會產生一系列的凝視與掃視反應，因此若能掌握眼球運動視覺軌跡，即能分析使用者之行為模式與認知學習歷程，而此模式已廣泛應用於各個領域之中。
過去所使用的注視追蹤方法，在蒐集注視數據資料時，通常會將使用者頭部固定，再進行注視模型訓練與分析，藉此提高訓練分類之準確率。然而當使用者頭部偏移時，則會導致注視分類預測之準確率降低，因此本研究探討非固定頭部的分類準確度。
本研究使用一般的網路攝影機，為了提升非固定頭部分類之準確度，過往的注視追蹤之研究常以眼睛外觀模型劃分注視區域，本研究則探討訓練模型架構結合卷積神經網路架構與循環神經網路之演算法，透過計算頭部姿勢預估中的俯仰角、偏航角與翻滾角加入模型訓練，使得使用者頭部能在偏移範圍於俯仰角+/-10°與偏航角+/-20°內移動，並且同時參考前一秒時間空間序列上的視線區域，再做注視點預測與分析，提高注視區域分類準確率表現。
透過本研究所提出CNN+RNN之訓練模型，在不同注視區域劃分下為2x2準確率達 98%、3x3準確率達 97%、4x4準確率達 90%、5x5準確率達 85%、6x6準確率達 80%、7x7準確率達 74%、8x8準確率達 69%、9x9準確率達 62%，相較於單一採用CNN架構訓練模型分類準確率，CNN+RNN模型架構能有效提升整體注視區域分類準確率 7~15%。

Eye trackers can accurately measure the user’s eye movement, trajectory, dilation, and constriction of the pupil. However, dedicated devices are expensive. This research looks at the use of web cameras to track and predict the gaze area. In particular, this study focuses on the accuracy of classification with some degree of freedom on head movement.

In this research, we propose a training model that uses convolution neural network and recurrent neural network in succession to train gaze direction. The user’s head can deviate within +/-10°pitch angle and +/-20° yaw angle. Regarding the gaze area of Space-Time Series of one second before, predicting and analyzing the gaze point, then improve the accuracy of classification of the gaze area. Experimental results show that the accuracy of automatic gaze tracking ranges between 98% to 62% as the gaze area gradually decreases from 640×360 pixels to 142×80 pixels.

[1] A. Fornaser, M. De Cecco, M. Leuci, N. Conci, M. Daldoss, A. Armanini, L. Maule1, F. D. N. &M. D. L. (2017). Eye tracker uncertainty analysis and modelling in real time. In Journal of Physics: Conference Series IOP Publishing., Vol. 778, No. 1, pp. 012002.
[2] Bacivarov, I., Ionita, M., &Corcoran, P. (2008). Statistical models of appearance for eye tracking and eye-blink detection and measurement. IEEE Transactions on Consumer Electronics, Vol. 54, No. 3, pp. 1312-1320.
[3] Blignaut, P. (2014). Mapping the pupil-glint vector to gaze coordinates in a simple video-based eye tracker. Journal of Eye Movement Research, Vol. 7, No. 1, pp. 1-11.
[4] Cherif, Z. R., Naït-Ali, A., Motsch, J. F., &Krebs, M. O. (2002). An adaptive calibration of an infrared light device used for gaze tracking. Conference Record - IEEE Instrumentation and Measurement Technology Conference, Vol. 2, pp. 1029-1033
[5] Hennessey, C., Noureddin, B., &Lawrence, P. (2006). A single camera eye-gaze tracking system with free head motion. In Proceedings of the 2006 Symposium on Eye Tracking Research & Applications, Vol. 1,pp. 87–94.
[6] Huang, J.Bin, Cai, Q., Liu, Z., Ahuja, N., &Zhang, Z. (2014). Towards accurate and robust cross-ratio based gaze trackers through learning from simulation. Eye Tracking Research and Applications Symposium (ETRA), pp. 75–82.
[7] Kar, A., &Corcoran, P. (2017). A review and analysis of eye-gaze estimation systems, algorithms and performance evaluation methods in consumer platforms. In IEEE Access, Vol. 5, pp. 16495-16519.
[8] King, D. E. (2009). Dlib-ml: A machine learning toolkit. Journal of Machine Learning Research, Vol. 10, pp. 1755-1758.
[9] Koutras, P., &Maragos, P. (2015). Estimation of eye gaze direction angles based on active appearance models. Proceedings - International Conference on Image Processing( ICIP), pp. 2424–2428.
[10] Meyer, A., Bõhme, M., Martinetz, T., &Barth, E. (2006). A single-camera remote eye tracker. Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol 4021, pp. 208–211.
[11] Murphy-Chutorian, E., &Trivedi, M. M. (2009). Head pose estimation in computer vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 31, No. 4, pp. 607-626.
[12] Oeda, S., Kurimoto, I., &Ichimura, T. (2006). Time series data classification using recurrent neural network with ensemble learning. Lecture Notes in Computer Science, Vol. 4253, pp. 742–748.
[13] Simonyan, K., &Zisserman, A. (2015). Very deep convolutional networks for large-scale image recognition. In The International Conference on Learning Representations (ICLR).
[14] Yoo, D. H., &Chung, M. J. (2005). A novel non-intrusive eye gaze estimation using cross-ratio under large head motion. Computer Vision and Image Understanding, Vol. 98, No. 1, pp. 25-51.
[15] Zhang, Z., &Cai, Q. (2014). Improving cross-ratio-based eye tracking techniques by leveraging the binocular fixation constraint. In Proceedings of the Symposium on Eye Tracking Research and Applications, Vol. 14, pp. 267–270.
[16] 劉文義. (1996). 以眼球電信號控制之殘障者人機介面設計. 國立台灣大學電機工程研究所碩士論文.
[17] 吳昭容. (2019). 眼球追蹤技術在幾何教育的應用與限制. 臺灣數學教育期刊, 6(2), 1–25.
[18] 林瑞硯. (2011). 使用網路攝影機即時人眼偵測與注視點分析. 臺灣師範大學碩士論文.
[19] 簡菁怡. (2009). 以彩色影像辨識為基礎之眼控系統研究與應用. 南台科技大學碩士論文.
[20] 葉俊材. (2011). 利用主動外觀模型估計眼睛注視方向. 國立臺灣科技大學碩士論文.
[21] 蔡金源. (1997). 以眼球控制之殘障者人機介面系統：紅外線視動滑鼠. 國立台灣大學電機工程研究所碩士論文.
[22] 許雅淳. (2013). 使用單一網路攝影機之視線判斷. 臺灣師範大學碩士論文.
[23] 賴孟龍, &陳彥樺. (2012). 以眼動方法探究幼兒閱讀繪本時的注意力偏好. 幼兒教保研究期刊, 8, 81–96.
[24] 辛孟錩. (2013). 基於影像結構相似性指標的單攝影機視線追蹤系統與應用. 國立東華大學碩士論文.
[25] 陳美琪. (2014). 基於二階層式支持向量機之即時注視區域分析. 臺灣師範大學碩士論文.