以穿戴式慣性感測器減低人體姿態估測搜尋空間之維度

Dimension Reduction of the Human-Posture-Estimation Search Space Using Wearable Inertial Sensors

許迺赫(Nai-He 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(2010年)

https://doi.org/10.6837/NCNU.2010.00305

摘要

本論文的主旨在於探討利用慣性感測器 (Inertial Sensor) 之特性協助人體姿態估測系統提升估測準確率以及減低搜尋空間維度之可行性。在本研究中我們建置一使用單攝影機以及穿戴式慣性感測器作為量測裝置，並利用粒子濾波器 (Particle Filter) 追蹤人體各部位方位之三維人體姿態估測系統。經由慣性感測器量測值推估人體姿態運動學限制，以達到縮小粒子濾波器搜尋範圍之成效。另外於本研究中也建置一套基於雙攝影機之簡易三維動作擷取系統 (MoCap)，利用穿戴於關節之紅外線標記重建其三維空間角度資訊做為參考資料來源。最後我們將呈現利用MoCap取得之參考資料位於慣性感測器訊號所推算出的搜尋空間中之實際分佈情形。實驗結果顯示，由MoCap所重建出的三維關節角度於可接受之誤差範圍內與我們所推估之運動限制模型吻合，因此我們即可將此運動限制套用於粒子濾波器以降低搜尋空間之維度，並加速姿態估測之計算速度。

關鍵字

穿戴式慣性感測器 ； 加速規 ； 人體姿態估測 ； 粒子濾波器 ； 動作捕捉器 ； 降維

並列摘要

This thesis aims to study the feasibility of using inertial sensors to improve the accuracy and to reduce the dimension of the search space in posture estimation. A three dimensional human posture estimation system is constructed which utilizes the particle filter technique to track the position and orientation of each body part using measurements from a monocular camera and
wearable inertial sensors. It is shown that the search space of the particle filter can be effectively reduced to a 1-D curve using kinematic constraints derived from the measurements of the inertial sensors. To verify the kinematic constraints, a simple motion capture (MoCap) system is constructed using the stereo vision technique. Infrared makers attached to joints are used to estimate the joint angles. The estimated joint angles are compared with the feasible solutions computed from the kinematic constraints. The experimental results show that the MoCap joint angles are consistent with the proposed kinematic constraints. Therefore, the kinematic constraints can be used to reduce the dimension of search space of the particle filter and to improve the system performance.

並列關鍵字

wearable inertial sensors ； accelerometer ； human pose estimation ； particle filter ； MoCap ； dimension reduction

參考文獻

[1] D. Ramanan and D. Forsyth, “Finding and Tracking People From the Bottom Up,” in 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2003. Proceedings, 2003.

[2] R. Navaratnam, A. Thayananthan, P. Torr, and R. Cipolla, “Hierarchical Part-based Human Body Pose Estimation,” in Proc. British Machine Vision Conference, Citeseer, 2005.

[3] K. Kunze, M. Barry, E. Heinz, P. Lukowicz, D. Majoe, and J. Gutknecht, “Towards Recognizing Tai Chi–An Initial Experiment Using Wearable Sensors,” Proc. Fifth International Forum on Applied Wearable Computing, 2006.

[4] J. Favre, R. Aissaoui, B. Jolles, O. Siegrist, J. de Guise, and K. Aminian, “3D Joint Rotation Measurement Using MEMS Inertial Sensors: Application to the Knee Joint,” ISB-3D: 3-D Analysis of Human Movement, 28-30 June 2006, Valenciennes, France, 2006.

[5] K. Kawano, S. Kobashi, M. Yagi, K. Kondo, S. Yoshiya, and Y. Hata, “Analyzing 3D Knee Kinematics Using Accelerometers, Gyroscopes and Magnetometers,” in IEEE International Conference on System of Systems Engineering, 2007. SoSE’07, pp. 1–6, 2007.