臺灣博碩士論文加值系統

近年來智慧型手機普及率日益增加，除了因為通訊和網際網路的使用需求外，由於手機上也較一般桌上型電腦與筆記型電腦多了一些感應器（Sensor），例如紅外線、GPS、藍芽、三軸加速器和陀螺儀等等，使得手機也有更多生活上的便利應用，例如以手機作為家電的遙控器或作為跌倒偵測的感知裝置等等，可以預見未來因為感測技術的快速發展，必然會有越來越多新穎的應用應運而生。除了手機之外，目前正廣受大家所注意的穿戴式裝置(Wearable Devices)，其實也是集各種感應器和聯網通訊功能於一身的隨身配件，因此利用各種感應器的資料來提供個人客製化的生活服務必是未來穿戴計算（Wearable Computing）中的主要範疇。基於此因，本研究針對三軸加速器(Triaxial Accelerometers)以及三軸陀螺儀(Triaxial Gyroscope)在空間中移動的軌跡提出了一套辨識技術，並利用此技術設計出分離數字辨識系統，讓使用者可持著智慧性手機在空中書寫數字來實施操控遠端設備或輸入手機認證密碼。有別於過往的以觸控面板輸寫文字或手勢，使用空中書寫在視覺的程度上擁有較高的保密性，且可不被限制在觸控的面板上，使用者也可在不解鎖手機螢幕的情況下直接達到更多操作效果。
在辨識的部分，我們主要改良動態時軸扭曲（Dynamic Time Warping, DTW）作為辨識方式，與DTW不同之處在於我們加入了自動化分群(Automatic Clustering)與統計模型來訓練資料，讓辨識系統可不需大量儲存比對樣板就可因應不同人的書寫習慣。且因應書寫軌跡的速度差異，與動態時軸扭曲在序列長度過大時的效能問題，我們在資料前處理的部分加入了自動序列濃縮(Automatic Segment Compression)，此濃縮技術大幅降低比對時間，而且也減少因為書寫快慢造成資料長短變化過大的影響，故能有效提昇辨識率，從實驗結果顯示，本論文提出的方法可以達到97.9%的辨識率，較其他方法更為準確。

In recent years, sensor in smart phone has more applications because the
popularity increasing, ex Remote Control, Fall Detection, etc. In addition to mobile
phones, Wearable Devices is also a portable accessory with sensors and networking.
Therefore, using sensor data to provide living services will be key of Wearable
Computing in the future. For this reason, this paper presents a handwritten recognition
technology based on Triaxial Accelerometers and Triaxial Gyroscope, and design the
discrete numeral recognition system. It allows users to write numerals in the air, to
control device or enter the password. Unlike in the past to touch screen or gestures, it
has higher confidentiality on the visual, and not be limited to the touch screen, users
can also achieve more operationally effect without unlocking.
In recognition, we mainly improved Dynamic Time Warping as recognition
function, and the difference is added Automatic Clustering and statistical models for
training data in response to different user writing habits without a lot of comparing
models. Because the write speed difference and Dynamic Time Warping performance
issues in sequence length too large, we incorporate an Automatic Sequence
Compression, to significantly reduce recognition time, and effectively improve the
recognition rate because reduce the impact of writing speed.
The experimental results show that the proposed in this paper achieves a
recognition rate 97.8%, it is more accurate than other methods.

目錄
誌謝 I
摘要 II
Abstract III
第1章 序論 1
1.1 研究動機 1
1.2 相關研究 2
1.2.1 感應器 2
1.2.2 軌跡辨識 2
1.2.3 動態時軸扭曲（Dynamic Time Warping，DTW） 3
1.2.4 隱馬爾可夫模型(Hidden Markov model，HMM) 4
1.3 系統流程 5
第2章 特徵抽取 7
2.1 硬體設備 7
2.2 數據正規化 9
2.3 自動化序列濃縮 10
第3章 數字辨識 20
3.1 動態時軸扭曲(Dynamic Time Warping，DTW) 20
3.2 自動化分群 21
3.3 樣板統計模型 24
3.3.1 建立模型 24
3.3.2 統計模型辨識 26
第4章 實驗結果 28
4.1 序列濃縮實驗 30
4.2 統計模型實驗 33
4.3 陀螺儀數據實驗 36
4.4 特定使用者與不特定使用者效果比較 37
4.5 與其他方法比較 38
第5章 結論與未來展望 39
5.1 結論 39
5.2 未來展望 40

[1] “維基百科 慣性導航系統,” [線上]. Available: https://zh.wikipedia.org/wiki/%E6%83%AF%E6%80%A7%E5%AF%BC%E8%88%AA%E7%B3%BB%E7%BB%9F.
[2] Bourke, Alan K., and Gerald M. Lyons, “A threshold-based fall-detection algorithm using a bi-axial gyroscope sensor,” Medical engineering &; physics, pp. 84-90, 2008.
[3] 林顗德, “在智慧型手機上應用重力感測器進行人體動作分析,” 淡江大學資訊工程學系, 2011.
[4] Bao, Ling, and Stephen S. Intille., “Activity recognition from user-annotated acceleration data,” Pervasive computing. Springer Berlin Heidelberg, pp. 1-17, 2004.
[5] Brezmes, Tomas, Juan-Luis Gorricho, and Josep Cotrina., “Activity recognition from accelerometer data on a mobile phone,” Distributed computing, artificial intelligence, bioinformatics, soft computing, and ambient assisted living. Springer Berlin Heidelberg, pp. 796-799, 2009.
[6] 伍星翰, “線性加速度感測器的研究與其應用,” 國立中央大學, 2009.
[7] 陳建榮, “基於智慧型手機之跌倒偵測與三維計步器之設計與實現,” 淡江大學資訊工程學系, 2014.
[8] Vail, Douglas, and Manuela Veloso., “Learning from accelerometer data on a legged robot,” 於 Proceedings of the 5th IFAC/EURON symposium on intelligent autonomous vehicles, 2004.
[9] Zhang, Shiqi, Chun Yuan, and Yan Zhang., “Handwritten character recognition using orientation quantization based on 3D accelerometer,” 於 Proceedings of the 5th Annual International Conference on Mobile and Ubiquitous Systems: Computing, Networking, and Services. ICST, 2008.
[10] Niezen, Gerrit, and Gerhard P. Hancke., “Gesture recognition as ubiquitous input for mobile phones,” 於 International Workshop on Devices that Alter Perception, 2008.
[11] Choi, Eun-Seok, et al., “Beatbox music phone: gesture-based interactive mobile phone using a tri-axis accelerometer.,” 於 Industrial Technology, 2005.
[12] Zhu, Chun, and Weihua Sheng., “Online hand gesture recognition using neural network based segmentation,” 於 Intelligent Robots and Systems, 2009.
[13] Choi, Sung-Do, Alexander S. Lee, and Soo-Young Lee, “On-line handwritten character recognition with 3D accelerometer,” 於 Information Acquisition, 2006.
[14] 王任宏, “以動態時軸扭曲進行連續手寫軌跡分段及辨識,” 國立東華資訊工程學系, 2009.
[15] 陳柏叡, “一筆劃軌跡中連續數字辨識,” 國立東華大學資訊工程學系, 2014.
[16] 周詠詩, “植基於動態規劃自動分群與時間扭曲之關鍵動作片段檢索,” 國立東華大學資訊工程學系, 2014.
[17] Keogh, Eamonn J., and Michael J. Pazzani., “Scaling up dynamic time warping for datamining applications.,” 於 Proceedings of the sixth ACM SIGKDD international conference on Knowledge discovery and data mining, 2000.
[18] Rath, Toni M., and Raghavan Manmatha, “Word image matching using dynamic time warping,” 於 Computer Vision and Pattern Recognition, 2003.
[19] Yamato, Junji, Jun Ohya, and Kenichiro Ishii, “Recognizing human action in time-sequential images using hidden markov model.,” 於 Computer Vision and Pattern Recognition, 1992.
[20] Bahl, L. R., et al., “Maximum mutual information estimation of hidden Markov model parameters for speech recognition,” 於 proc. icassp, 1986.
[21] Schuller, Björn, Gerhard Rigoll, and Manfred Lang, “Hidden Markov model-based speech emotion recognition.,” 於 Acoustics, Speech, and Signal Processing, 2003.
[22] “Hidden Markov Model (HMM) Toolbox for Matlab,” [線上]. Available: http://www.cs.ubc.ca/~murphyk/Software/HMM/hmm.html.