基於特徵提取的低功耗應用人體即時姿態識別系統

A Low Cost Human Posture Recognition Based on Feature Extraction for Real-Time Applications

蔡易澂(I-Cheng Tsai)

指導教授 ： 邱瀞德

國立清華大學/電機資訊學院/通訊工程研究所/碩士(2012年)

摘要

人體動作的辨識在很多應用中扮演了一個很重要的角色，像是視
訊監視系統、醫療分析等等。因此，很多研究學者提出關於人體姿勢
和行為的辨識的研究，像是利用一個姿勢的骨幹當做特徵，並且利用
隱馬爾可夫模型來辨識其姿態。其他像利用一個姿態經過三角化後並
算出其個別三角型的中央質心，然後再建立其骨幹並分析其姿態。前
者在需要大量的訓練樣本，後者的演算法需要很高的計算複雜度。
在這篇論文裡，我們提出一個以中央質心為基礎的姿勢和行為的
分析。在這裏我們只用了五個重心點和四個特徵集合來做分析，其中
兩個特徵用來量測重心點在垂直跟水平方向的變化，另外兩個用來量
測人體上半身和下半身角度變化的速率。根據這些定義的特徵集合和
我們提出的分類模型，我們提出的方法可以辨識出五種不同的靜態姿
勢，像是站、坐、蹲、躺和彎腰，且也可以分出兩種動態姿勢，像是
揮手和走路。這實驗結果顯示我們提出的方法比起其他的方法有好的
辨識能力和較小的計算複雜度。

關鍵字

姿勢辨識 ； 行為分析 ； 重心 ； 影像監視

並列摘要

The recognition of human actions plays an essential role in many applications such as hu-
man machine interaction, surveillance systems, medical data analysis and etc. With the
popularity of smart televisions and handheld devices, low cost and real-time realization
of human posture recognition become important. Many researches have been proposed
for human posture and action recognition. However, most of the proposed approaches
have high computational complexity. In this paper, we propose a low-cost real-time ac-
tion recognition approach using only ve center of gravity points (COG) and four feature
sets. Two feature sets measure the displacement of the upper and lower body COG in
the vertical and horizontal directions. The other two feature sets quantize the upper
and lower body angular change rate. With these feature sets and a classication model,
our proposed approach is able to recognize ve dierent static postures including stand,
laying, bend, sit and squat and two actions, walking and handwaving. The simulation
results show that our proposed approach achieve 98.02% to 80.20% recognition rates for
various postures and actions in the KTH and ISIR databases. Our approach can achieve
real-time recognition for video sequence and has lower computational complexity than
other state-of-art algorithms.

並列關鍵字

posture recognition ； behavior analsis ； center of gravity ； video surveillance

參考文獻

[1] T. B. Moeslund and E. Granum, A survey of computer vision-based human motion

capture," Comput. Vis. Image Underst., vol. 81, no. 3,pp. 231-268, Mar. 2001.

[2] W. Hu, T. Tan, L.Wang, and S.Maybank, A survey on visual surveillance of object-

[3] D. M. Gavrila, The visual analysis of human movement: A survey, " Comput. Vis.

via feature-aided tracking and hidden Markov models," IEEE Trans. Syst. , Man,