人體骨架模型於姿態辨識應用之研究

擷取人體動作資訊在電腦視覺中，有許多重要的應用，例如:姿態辨識、互動遊戲、虛擬實境、居家照護等。雖然目前已有許多動作辨識的方法被提出，但由於人體具有非剛性特性，要能準確辨識動作姿態，仍充滿著許多的困難與挑戰。利用電腦視覺從拍攝畫面中，分析使用者的動作變化與姿態，具有方便、非侵入與低成本的優點，一直是姿態辨識的研究重點。由於攝影機所擷取的影像資料易受光影變化、雜訊干擾等因素的影響，如何克服這些干擾，同時準確辨識不同的姿態的動作，是應用電腦視覺進行人體姿態辨識的關鍵技術。本研究旨在應用骨架模型建立一套以電腦視覺為基礎的姿態辨識系統。本研究除分別探討人體骨架模型與星狀骨架模型於姿態辨識應用外，並實際以四種姿態做測試：雙手平放、左右側彎、雙手高舉和雙手上下擺動。星狀骨架模型是根據物件中心連接到物件輪廓突出點的方法，而人體骨架模型是針對各部位做偵測所組合成的。骨架模型係針對人體姿勢以代表性的描述，識別各動作之特徵，再以隱藏馬可夫模型，進行姿態辨識。由實驗結果顯示，整體姿態辨識率達93%，人體各部位偵測率平均達82%，證明本文提出之方法可有效辨識使用者動作。

關鍵字

姿態辨識；隱藏式馬可夫模型；星狀骨架模型；人體骨架模型

並列摘要

There are many important applications for capturing human movement information in the computer vision field, such as posture recognitions, interactive games, virtual reality and home care services. Although many approaches of posture recognition have been proposed so far, to accurately recognize posture is still a challenge because of the human body's non-rigid characteristics. Analyzing the human movement information and posture from capturing video streams through computer vision technique is convenient, non-invasive and at low cost, which has always been an active research topic in posture recognition. Since the images captured by camera is easily influenced by the change of light, noise interference and other factors, the critical techniques for recognizing people posture adopting computer vision skills is to overcome these disturbances and recognize different kinds of posture simultaneously. This thesis proposes a skeleton model adopted to develop a computer vision based posture recognition system. Due to the differences between human skeleton model and star skeleton model applied on posture recognition, four types of postures are utilized in the test: two-hands down, side bent, two-hands up and hand-waving movements. The star skeleton model represents the human body through connections between the center and other salient points, whereas the human skeleton model considers the body as the combination of individual human body spots. The skeleton model represents each human body spots as graphic descriptions, and transforms to features recognized by hidden Markov model. Experimental result shows that the average recognition rate is 93%, and the average of each body part’s detection rate is 82%; therefore, proving the efficiency of the four types of posture that the user poses.

並列關鍵字

Posture Recognition ； Hidden Markov Model ； Star Skeleton Model ； Human Skeleton Model

參考文獻

[1] J. W. Davis and A. F. Bobick, "The representation and recognition of human movement using temporal templates," IEEE Conference on Computer Vision and Pattern Recognition, 1997, pp.928-934.

[2] G. Shakhnarowich, P. Viola and T. Darrell, "Fast pose estimation with parameter-sensitive hashing," International Conference on Computer Vision, 2003, pp.750-757.

[3] P. F. Felzenszwalb, "Pictorial structures for object recognition," International Journal of Computer Vision, vol. 61, no. 1, 2005, pp.55-79.

[4] S. Wang, H. Ai, T. Yamashita and S. Lao, "Combined top-down/bottom-up human articulated pose estimation using adaboost learning," International Conference on Pattern Recognition, 2010, pp.3670-3673.

[5] N. Dalal and B. Triggs, "Histograms of oriented gradients for human detection," IEEE Conference on Computer Vision and Pattern Recognition, 2005, pp.886-893.

國際替代計量

人體骨架模型於姿態辨識應用之研究

未授權

主題瀏覽