近幾年來，穿戴式裝置漸漸地成為世界上的主流，腦機介面(Brain-Computer
Interface, BCI)就是其中之一，它是測量腦頭皮電位(electroencephalography ,EEG)
的訊號，對使用者而言，腦頭皮電位的取得也越來越容易和便宜。腦機介面通常
的使用在醫療還有娛樂上。為了更便利的取得腦頭皮電位，我們使用了無線可攜
帶式的儀器，因為關係的無線傳輸的問題，在此篇論文中，我們引用了[1]的壓
縮技術和眼動偵測，並且針對個別使用者進行訓練。 首先，我們使用了EPOC
的腦波量測儀器去取得腦頭皮電位(EEG)的訊號，並且Emotiv 提供一個連結電腦
和儀器的介面。為了可以偵測八個方向和找出每個使用者的參數。在這篇論文中，
我們提出了兩種模式，分別是訓練模式(training mode) 和 偵測模式(detection
mod)。 在訓練模式中，我們儲存一段只有左右和上下眼動的腦頭皮電位(EEG)訊
號。並使用獨立分項分析(Independent Component Analysis , ICA)去萃取眼動
特徵。因為獨立分項分析是屬於未知原始訊號的演算法(blind source algorithm)，
所以我們利用權重選擇(weighting selection)去找出左右和上下眼動的權重，並且
也找出左右和上下眼動的最大值最小值。訓練者模式結束後，我們會得到左右和
上下眼動的權重以及最大值最小值，將這些數值使用在偵測模式中去偵測八個方
向。我們使用了隱藏馬可夫模型(Hidden Markov Model , HMM)去進行眼動的分類，
再將左右和上下的分類結果結合成八個方向。最後我們可以使用八個方向去控制
滑鼠點擊螢幕小鍵盤並且在文件中輸入字母。

In recent years, the wearable devices is an essential trend of world, among them brain-
computer interface(BCI) utilized electroencephalography (EEG) signal measurement be-
comes more accessible and cheaper for users. BCI is commonly used for medical usage
or entertainment applications. For EEG signal acquisition, wireless portable devices
are preferred due to their convenience. In the wireless scenario, power and bandwidth
become important issues. In this thesis, we follow the compress technique and detection
mode in [1], and provide independent training for each user. At rst, we used EPOC
headset to acquire EEG signals, and Emotiv oer API between EPOC device and PC.
In order to detect eight direction eye movement and nd parameter for each user. In
this thesis, we proposed two mode in our system, respectively training mode and de-
tection mode. In training mode, we store the EEG signal that only Left/Right and
Up/Down eye movement, and used Independent Component Analysis(ICA) to extract
feature. Because of ICA is blind source algorithm, we used weighting selection to nd
Left/Right and Up/Down weighting, and nd Max/Min value of eye movement signal.
After training mode, we used Left/Right and Up/Down weighting and Max/Min value
of eye movement signal to detect eight direction in real-time. We used Hidden Markov
Model (HMM) to do eye movement direction classication, and combined Left/Right
and Up/Down result to detect eight direction. Finally we can control eight direction of
mouse to click monitor keyboard to key the words in the text le.

[1] H.-P. Chu, \A compressive sensing aided eeg-based bci system for eye movement
direction classication using hmm," October 2013.
[2] T. Lee, \Independent component analysis: Theory and appliciation," Kluwer Aca-
demic Publishers, Boston ISBN : 0-7923-8261-7, pp. 7{11; 18{21; 27{49; 83{100,
1998.
[3] \http://www.cqasc.co.uk/."
[4] J. J. Vidal, \Toward direct brain-computer communication," Annual Review of
Biophysics and Bioengineering, vol. 2, pp. 157{180, June 1973.
[5] A. Funase, T. Yagi, Y. Kuno, and Y. Uchikawa, \Prediction of eye movements from
EEG," in International Conference on Neural Information Processing, vol. 3, 1999,
pp. 1127{1131.
[6] A. Funase, T. Hashimoto, T. Yagi, A. K. Barros, A. Cichocki, and I. Takumi, \Re-
search for estimating direction of saccadic eye movements by single trial processing,"
in IEEE Engineering in Medicine and Biology Society, 2007, pp. 4723{4726.
[7] T. Ito, T. Shinji, H. Sumiya, and M. Baba, \Eye movement-related EEG potential
pattern recognition for real-time bmi," in Proceedings of SICE Annual Conference,
2010, pp. 1055{1059.
60 BIBLIOGRAPHY
[8] S. S. Gupta, S. Soman, P. G. Raj, R. Prakash, S. Sailaja, and R. Borgohain,
\Detecting eye movements in EEG for controlling devices," in IEEE Conference on
Computational Intelligence and Cybernetics, 2012, pp. 69{73.
[9] A. Cohen, \Hidden markov models in biomedical signal processing," in IEEE Con-
ference Engineering in Medicine and Biology Society, vol. 3, 1998, pp. 1145{1150.
[10] E. J. Cands and M. B. Wakin, \An introduction to compressive sampling," IEEE
Signal Processing Magazine, vol. 25, no. 2, pp. 21{30, 2008.
[11] BCI-2000, \http://www.bci2000.org/wiki/index.php/user tutorial:eeg measurement setup,"
September 2012.
[12] sccn, \sccn.ucsd.edu/eeglab/."
[13] T.-P. Jung, S. Makeig, M. Westereld, J. Townsend, E. Courchesne, and T. J. Se-
jnowski, \Independent component analysis of single-trial event-related potentials,"
in Proc. ICA', Jan. 1999, pp. 173{178.
[14] H. Yu, H. Lu, T. Ouyang, H. Liu, and B. Lu, \Vigilance detection based on sparse
representation of EEG," in Proc. IEEE EMBC, 2010, pp. 2439{2442.
[15] A. J. Bell and T. J. Sejnowski, \An information-maximization approach to blind
separation and blind deconvolution," Neural Computation, vol. 7, pp. 1129{1159,
1995.
[16] T. W. Lee, M. Girolami, and T. J. Sejnowski, \Independent component analysis
using an extended infomax algorithm for mixed sub-gaussian and super-gaussian
sources," Neural Computation, vol. 11, no. 2, pp. 409{433, 1999.
[17] S. B. A. J. T-P and T. Sejnowski, \Independent component analysis of electroen-
cephalographic data," D. Touretzky M. Mozer and M. Hasselmo (Eds). Advances
BIBLIOGRAPHY 61
in Neural Information Processing Systems,MIT Press, Cambridge, pp. 8:145{151,
1996.
[18] P. Blunsom, \Hidden markov models," 2004.
[19] G. D. Forney and Jr., \The viterbi algorithm," Proceedings of the IEEE, vol. 61,
no. 3, pp. 268{278, 1973.
[20] S. Chatterjee, K. Hari, P. Handel, and M. Skoglund, \Projection-based atom selec-
tion in orthogonal matching pursuit for compressive sensing," in National Confer-
ence on Communications, 2012, pp. 1{5.
[21] J. A. Tropp and A. C. Gilbert, \Signal recovery from random measurements via
orthogonal matching pursuit," Information Theory, IEEE Transactions on, vol. 53,
no. 12, pp. 4655{4666, dec. 2007.
[22] A. M. Abdulghani, A. J. Casson, and E. Rodriguez-Villegas, \Quantifying the
performance of compressive sensing on scalp EEG signals," in Proc. ISABEL, 2010,
pp. 1{5.