以基於Kinect的特徵點偵測實現實時手勢估測

Real-time Hand Pose Estimation with Feature Points Detection using Kinect

張哲豪(Che-Hao Chang)

指導教授 ： 黃仲陵 ； 林嘉文

國立清華大學/電機資訊學院/電機工程學系/碩士(2013年)

https://doi.org/10.6843/NTHU.2013.00063

摘要

本論文提出了一種基於深度影像來實現快速且準確的手勢參數估測系統。此系統先將深度影像轉換為特徵向量，再以回歸函式得出手指關節參數。有別於目前手勢估測的兩種主要方向Model-based和Appearance-based，此方法只需要極短的時間便可計算出值域連續的參數。一開始先從深度影像中擷取出手部影像並標準化，透過隨機森林分類器找出手上的幾個特定點，將其轉換為不受手部旋轉影響的特徵向量之後，再使用預先訓練好的回歸函式計算出手勢參數。其中我們所使用的手部影像擷取系統可以不受手部旋轉影響而準確地定位出手掌中心，標準化使得手部的大小不隨距離遠近而變化。隨機森林分類器中所使用的分裂函式經過我們的修改後，得以處裡在影像平面上的旋轉情況，使其更加適合應用在手勢偵測上。實驗方面，我們將提出的方法所估算的參數與電子手套取得的實際參數做比對，來計算此系統的準確度。我們也統計出不同距離以及不同的手部旋轉角度對準確度的影響；並且，不同的測試者對系統的影響與其應對方法也在文章的最後一併討論。

關鍵字

手勢估測 ； 回歸函數 ； 隨機森林 ； 深度影像

並列摘要

This thesis presents a real-time and precise depth image based hand pose estimation method. The depth image obtained from Kinect is converted into a feature vector for regression functions to retrieve hand joint parameters. Different from the two mainly proposed methods, model-based and appearance-based, our approach retrieves continuous result within a short period of time. In the beginning, the hand region is segmented from the depth image. Some specific feature points on the hand are located by random forest classifier, and the relative displacements of these feature points would be transformed into a rotation invariant feature vector. Finally, the system retrieves the hand joint parameters by using regression functions which are trained off-line. The results of the proposed method are compared with the ground truth obtained by data glove to evaluate the system reliability. The effects of different distances and different rotation angles to the estimation accuracy are evaluated.

並列關鍵字

hand pose estimation ； regression function ； random forest ； depth image

參考文獻

[16] C. C. Chung and L. C. Jen, "LIBSVM : a library for support vector machines," ACM Transactions on Intelligent Systems and Technology, pp. 27:1--27:27, 2011.

[1] M. Półrola and A. Wojciechowski, "Real-Time Hand Pose Estimation Using Classifiers," ICCVG, pp. 573-580, September 2012.

[2] H. Teleb and G. Chang, "Data glove integration with 3D virtual environments," Systems and Informatics (ICSAI), pp. 107 - 112, May 2012.

[3] R. Y. Wang and J. Popović, "Real-Time Hand-Tracking with a Color Glove," ACM SIGGRAPH, August 2009.

[5] J. Romero, H. Kjellström and D. Kragic, "Hands in action: real-time 3D reconstruction of hands in interaction with objects," Robotics and Automation (ICRA), pp. 458 - 463, May 2010.