結合微機電系統感測器與視覺感測器追蹤三維人體姿態

Combining MEMS and Visual Sensors for 3-D Human Posture Tracking

許瑞文(Jui-Wen Hsu)

指導教授 ： 石勝文

國立暨南國際大學/科技學院/資訊工程學系/碩士(2008年)

摘要

人體結構具有多個活動關節，這些關節的整體自由度非常高，所以要估測人體姿態的挑戰性也很高。在本論文中我們發展了一個有效率的人體姿態偵測系統。我們使用視覺感測器以及微機電感測器來取得姿態相關的資訊，利用人體關節角度資訊建立初始三維人體姿態模型，透過階層式的貝氏濾波器計算近似目標姿態的三維人體模型。其中微機電感測器提供的訊號可用來降低姿態偵測時的搜尋範圍，以增進其計算效率。在階層式貝氏濾波器中，我們推導了一個相似度函式來估算由視覺感測器及微機電感測器的量測值推估的人體模型與目標姿態的相似度。在實驗中，我們測試了數個運動序列，以証實本系統能夠提昇姿態估測運算速度並解決單由視訊無法解決的運動姿態估測問題。

關鍵字

追蹤三維人體姿態 ； 可穿戴式感應器 ； 加速規 ； 貝氏濾波器

並列摘要

Human body is an articulated object with high degrees of freedom and thus it is a great challenge to estimate the human posture. In this thesis, we develop an efficient human posture detection system which uses a visual sensor and micro electro-mechanical system (MEMS) sensors to obtain posture related information. Given initial values of human body joint angles, an initial 3D humanoid model can be established. The joint angles are refined using the hierarchical Bayesian filter approach. The measurements of the MEMS sensors can be used to reduce the searching space of the joint angles and thus the computation efficiency can be improved. To utilize the hierarchical Bayesian filter in posture estimation using both visual and MEMS sensors, we have derived a similarity function for estimating the similarity between the synthetic and the acquired real images of a human posture. In the experiment, we tested our system using several motion sequences verifying that the proposed method not only can improve the posture estimation efficiency but also can solve some problems which are very difficult when using visual sensors only.

並列關鍵字

Wearable Sensors ； Accelerometer ； Tracking Human Posture ； Bayesian filter

參考文獻

[1] H. A. W. Group, “H-Anim,” http://h-anim.org/Specifications/H-Anim1.1/h-anim1_1g.gif.

[2] I. Rius, D. Rowe, J. Gonzalez, and X. Roca, “A 3D dynamic model of human actions for probabilistic image tracking,” Pattern Recognition and Image Analysis, pp. 529–536, 2005.

[3] C. Curio and M. A. Giese, “Combining view-based and model-based tracking of articulated human movements,” in Proceedings of the IEEE Workshop on Motion and Video Computing, vol. 2, 2005, pp. 261–268.

[4] W. T. F. Z, N. R. Howe, N. R. Howe, M. E. Leventon, M. E. Leventon, and W. T. Freeman, “Bayesian reconstruction of 3D human motion from single-camera video,” in Advances in Neural Information Processing Systems, 1999, pp. 820–826.

[5] R. Cucchiara, C. Grana, A. Prati, and R. Vezzani, “Probabilistic posture classification for human-behavior analysis,” IEEE Transactions on Systems, Man and Cybernetics, Part A, vol. 35, no. 1, pp. 42–54, 2005.