改良型分層隱藏式馬可夫模型於手勢識別之研究

Gesture Recognition Using Improved Hierarchical Hidden Markov Model

林本煌(Ben-Huang Lin)

指導教授 ： 練光祐

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2011年)

https://doi.org/10.6841/NTUT.2011.00714

摘要

近年來科技發展日新月益，人機互動的介面不再被侷限，各式感測器都可成為人機互動的一部份。而手部動作是人類表達自我意圖的一種常見方式，因此利用手勢辨識進行人機互動的方式，在學術上是一個非常值得研究的議題。手勢識別的偵測方式可概分為影像處理或是使用感測器擷取特徵值兩大類，而影像方式雖可實現無穿戴的優點，但其定位擷取特徵困難且演算法繁複，因此手持式感測裝置仍為目前主流。本文乃採取手持式慣性感測器的方式來做手勢識別，擷取手部移動時產生的加速度變化量，傳輸至電腦儲存，再使用分層隱藏式馬可夫模型(HHMM)為主幹，結合手勢單位的概念，做為模型契合機率的辨別；而本文也將原始HHMM的Forward演算法加以改良，在不影響辨識率的條件下有效降低計算時間。原始演算程式在序列過多時，所需時間可能以小時為單位，而改良後之演算法則只需數分鐘即可完成計算。在本文實驗中以所有10個阿拉伯數字為辨識目標，在使用者手持慣性感測器揮動後，擷取加速度計中X、Y雙軸之數值，並在電腦中以演算法計算後，可辨識出使用者揮動輸入的阿拉伯數字，實驗結果顯示本辨識方法有極佳的正確性。

關鍵字

分層隱藏式馬可夫模型 ； 手勢識別 ； 手勢單位

並列摘要

Human-computer interaction issue becomes more and more important in recent years. Its application is not limited to KVM (keyboard, video, and mouse), but also includes microphones, cameras, and other various sensors. Since gesture is one of the most important expression methods for human intention, this motivates us to focus our research on gesture recognition. We believe it will be a powerful method for human-computer interaction. Although there are many detective methods of gesture recognition, such as vision-based approaches which can achieve the goal of gesture recognition without wearing any device, the feature extraction is indeed difficult and the algorithms are complex, too. Hence, handheld mobile devices are still popular at present. In this thesis, we present an improved method of gesture recognition using a handheld inertial sensor. The method starts from the treatment of acceleration signals coming from an inertial measurement unit (IMU). Our approach is developed based on hierarchical hidden Markov model (HHMM) and the concept of partitioning the gesture signal into small units. Then, the recognition is achieved by computing model likelihood probability. Here, we reduce computation time by improving “Forward Algorithm” of HHMM without affecting the recognition success rate. Finally, we set up the experiment environment to recognize all ten Arabic numerals. X- and Y-axis acceleration signals from a swing IMU are used as the input to the proposed algorithm. The results show very high accurate rate after a series of experiments have been carried out.

並列關鍵字

Hierarchical Hidden Markov Model ； Gesture Recognition ； Gesture Units

參考文獻

[5] C. Zhu, W. Sun, and W. Sheng, “Wearable Sensors based Human Intention Recognition in Smart Assisted Living Systems,” IEEE International Conference on Information and Automation, Changsha, 2008, pp. 954-959.

[6] C. Zhu, Q. Cheng, and W. Sheng, “Human Intention Recognition in Smart Assisted Living Systems Using A Hierarchical Hidden Markov Model,” IEEE International Conference on Automation Science and Engineering, Arlington, VA, 2008, pp. 253-258.

[8] 鍾哲民，加速度動作辨識系統之研究及應用，碩士論文，國立成功大學工程科學系，臺南，2008。

[13] Y. F. Zhang, Q. F. Zhang, and R. H. Yu, “Markov Property of Markov Chains and Its Test,” Proceedings of the Ninth International Conference on Machine Learning and Cybernetics, Qingdao, 11-14 July 2010, pp. 1864-1867.

[15] L. R. Rabiner, “A Tutorial on Hidden Markov Models and selected applications in speech recognition,” Proceedings of the IEEE, vol. 77, 1989, pp. 257-286.