利用簡易規則達成快速人體姿態鑑別

Fast Human Posture Recognition by Heuristic Rules

李乾丞(Chien-Cheng Li)

指導教授 ： 陳永耀

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2006年)

https://doi.org/10.6342/NTU.2006.10522

摘要

辨識人體運動行為是一項很有趣的研究主題，通常是使用電腦視覺來做為研究的方法，應用的範圍很廣泛，包含虛擬實境、人機介面、監控系統與人體動作分析。含有人體運動的連續畫面當中，人體運動行為之辨識可分為兩大部分:如何取得人體特徵的資訊以及人體動作行為之解析與表達。人體特徵取得的困難點在於要如何從拍攝取得的複雜影像中找出所需的人體部分，而人體動作行為解析與表達的挑戰在於當取得人體特徵之後，如何利用人體特徵進而辨識出多變且複雜的人體動作。
本篇論文即是針對人體動作行為的解析與表達這方面，提出一套適合居家看護的人體姿態辨識系統。其做法是先依據由CCD取得的人體輪廓，進而收集與整理人類姿態的共通點與特徵，再利用人體姿態的特徵來設計一些簡單的參數與規則，希望透過這些參數與簡易的規則可以反推出人體姿態，特別是針對一些日常生活中常見的姿態來做辨識，目的不僅是希望可正確的判斷出人體姿態，更希望可以提升辨識人體姿態的速度，達成真正即時系統，以便更符合居家看護系統的要求。本系統假設有ㄧ完整且清晰的人體輪廓做為輸入，因此本篇論文使用ㄧ個簡易的方法取得人體輪廓並且控制拍攝背景與穿著，再透過一些辨識規則辨認出人體姿態，在最後結論中將會顯示判斷正確率，以證明此方法可成功解析人體姿態。

關鍵字

人體姿態辨識 ； 電腦視覺 ； 居家看護系統

並列摘要

Recognition of human activities has become a popular research in recent years, especially in the field of computer vision. There are many applications, like virtual reality, human-computer interface, surveillance systems, and human actions analysis. There are two aspects about recognition of human activities: how to get the information of human activities and representation of human motion. The difficulties of gathering information of human activities are to extract human body characteristics from the complex image. The challenge of representing human motion is how to utilize the collected human information to recognize the complicated and varied human motions.
This thesis mainly focuses on the representation of human postures and supplies a methodology of human postures recognition for home-care system. This thesis assumes that the human silhouette would be got and the human silhouette is complete, though the background subtraction is used in this thesis. The methodology of human posture recognition is to search and create the relation between human silhouette and result of recognition. The relation of this thesis is parameters and simple rules. Parameters and rules mean the characteristics of human postures. To raise the correct recognition rate and to achieve a real time system are two goals in our research. In the later chapter of this thesis, there are some results to prove that these two goals are achieved in this thesis.

並列關鍵字

human posture recognition ； computer vision ； home-care system

參考文獻

[64]. 李冠德, “Fuzzy Rule-Based Human Actions Recognition for Home Care System”, 國立台灣大學電機工程研究所碩士論文, 2004.

[1]. I. Haritaoglu, D. Harwood, L. S. Davis, “W : real-time surveillance of people and their activities”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(8), 2000, pp.809-830.

[2]. N. M. Oliver, B. Rosario, and A. P. Pentland, “A Beyesian computer vision system for modeling human interactions”, IEEE transactions on Pattern Analysis and Machine Intelligence, 22(8), 2000, pp.831-843.

[3]. J.K. Aggarwal and Q. Cai, ”Human motion analysis: a review”, Nonrigid and Articulated Motion Workshop, 1997. Proceedings, IEEE, Page(s):90 – 102,16 June 1997.

[4]. G. Johansson, “Visual motion perception”, Sci. Am. 232(6), 1975, 76-88.