本論文提出了一個以patch為範例基準的近似法(example-based approach)來估測人體的運動關節參數。大多數的演算法用人體的整體姿勢特徵來當作範例，我們使用人體的剪影輪廓上萃取出來的patch影像來當我們的範例。即是說，以人物剪影的輪廓作為依據，並且使用local patches為單位與資料庫的patches作比對，來篩選出最近似於目前輸入畫面的人物的關節角度參數。
這個演算法的宗旨在於：當輸入畫面的人物影像的輪廓與資料庫裡的某個人體姿勢的輪廓相似的話，我們便當作輸入畫面的關節角度參數與此資料庫的姿勢的角度參數相同。所有patch的萃取方法是沿著人體影像的輪廓來萃取，並且記錄該patch範圍內的shape context。在辨識的時候，每一張輸入畫面都萃取數十個patch後，每一張patch都從資料庫中找出最鄰近的幾個patch，最後利用這些patch作相似度的投票，進而估測出最相像的輪廓而得到關節角度參數。為了加快最鄰近搜索的速度，我們使用LSH (Locality-Sensitive Hashing)來做相似patch搜尋的動作。它可以有效地降低相似資料搜尋的計算時間，並且不會因為資料庫資料量的增加而增加計算時間。最後，配合時間相依性與關節點預測來決定最後參數估測的結果。

Recently, estimating human poses from a monocular view has broad applications in human-computer interface, virtual reality and video surveillance. In this thesis, we present an example-based approach for 3D human body pose estimation from a static silhouette image. To improve the estimating accuracy of the occluded body parts, we use Microsoft kinect to capture the depth image of human model as our example features. Our main idea of estimation is that if the contour of the input human image is similar with the contour of the model in database, the pose parameters are considered to be the same. And to reduce the estimation time, we apply the Locality-Sensitive Hashing to index the global parameter of the human pose
First, we use the background model constructed by depth image to apply background subtraction and get the human silhouette. This helps to reduce the effect of light and shadow. After several morphological processes, we can get the human contour image. Then we use the shape context of the patches extracted from the contour image as the feature vector. After training state, we get a useful hash function to encode each feature vector into a hash value which is an index number for hash table construction. A useful hash function supposed to encode similar patches as same hash value and different patches as different hash values. At testing state, according to the hash value of input patch, we retrieve similar patches from the hash table and apply the Hough voting algorithm. After temporal and prediction constraint, the pose parameters with the highest voting are considered to be our estimation result.

參考文獻
[1] S. Belongie, J. Malik, and J. Puzicha. Shape Matching and Object Recognition Using Shape Contexts. PAMI, 24(4):509–522, 2002.
[2] X. Ren, A.C. Berg, and J. Malik, Recovering Human Body Configurations using Pairwise Constraints between Parts. ICCV, 2005.
[3] M.W. Lee and I. Cohen. A Model-Based Approach for Estimating Human 3D Poses in Static Images. PAMI, 28(6):905-916, 2006.
[4] L. Sigal and M.J. Black. Predicting 3D People from 2D Pictures. AMDO, 2006.
[5] G. Mori and J. Malik. Recovering 3D Human Body Configurations Using Shape Contexts. PAMI,28(7):1052-1062, 2006.
[6] A. Gionis, P. Indyk, and R. Motwani. Similarity Search in High Dimensions via Hashing. VLDB, 1999.
[7] G. Shakhnarovich, P. Viola, and T. Darrell. Fast Pose Estimation with Parameter Sensitive Hashing. ICCV, 2003.
[8] V. Athitsos, J. Alon, S. Sclaroff, and G. Kollios. BoostMap: An Embedding Method for Efficient Nearest Neighbor Retrieval. PAMI, 30(1):89-104, 2008.
[9] B. Leibe, A. Leonardis, and B. Schiele. Combined Object Categorization and Segmentation with an Implicit Shape Model. ECCV Workshop on Statistical Learning in Computer Vision, 2004.
[10] D. Demirdjian and R. Urtasun. Patch-Based Pose Inference with a Mixture of Density Estimators. AMFG, 2007.
[11] K. Hara. Real-time Inference of 3D Human Poses by Assembling Local Patches. WACV, 2009.
[12] Y. Cheng. Mean shift, mode seeking, and clustering. PAMI,17(8), 790-799(1995).
[13] C. Sminchisescu and B. Triggs. Kinematic Jump Processes for Monocular 3D Human Tracking. CVPR, 2003.
[14] R. Roppe, “Vision-based human motion analysis: An overview,” CVIU, vol. 108, 2007.
[15] Microsoft Corp. Remond WA, Kinect for Xbox 360.
[16] J. Shotton, A. Fitzibbon, M. Cook, T. Sharp, M. Finocchio, R. Moore, A. Kipman, and A. Blake, “Real-time Human Pose Recognition in Parts from Single Depth Images,” Pattern Recognition 2011.
[17] J. Deutscher, A. Blake, and I. Reid” Articulated Body Motion Capture by Annealed Particle Filtering” IEEE CVPR 2000.
[18] P. Peursum. “On the Behaviour of Body Tracking with the Annealed Particle Filter in Realistic Conditions” Technical Report of the Dept. of Computing, Curtin Univ. of Technology, 2006.
[19] X. Xu, and B. Li “Learning Motion Correlation for Tracking Articulated Human Body with a Rao-Blackwellised Particle Filter” IEEE ICCV 2007.
[20] J. Darby, B. Li and N. Costen “Tracking a Walking Person using Activity-Guided Annealed Particle Filtering” IEEE Int. Conf. on Automatic Face and Gesture Recog. 2008.
[21] J. Darby, B. Li and N. Costen “Behavior Based Particle Filtering for Human Articulated Motion Tracking”, ICPR 2008.