以深度資訊為基礎的寬鬆靜態手勢辨識; Depth-based Loose Static Gesture Recognition

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/61581

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/61581

Title:	以深度資訊為基礎的寬鬆靜態手勢辨識;Depth-based Loose Static Gesture Recognition
Authors:	劉書銘;Liu,Shu-ming
Contributors:	資訊工程學系
Keywords:	Kinect感應器;深度影像;寬鬆;靜態;手勢辨識;人機互動;Kinect;Depth Image;Loose;Static;Gesture Recognition;Human Computer Interaction
Date:	2013-08-26
Issue Date:	2013-10-08 15:22:53 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來，手勢辨識已成為重要的研究議題；廣泛應用在遊戲控制、家電操作、機器人控制等操作。然而，一些手勢辨識應用容易受到環境光源及背景影響，亦或是受限於端正標準手勢，其手指不彎曲，讓使用者無法舒適使用。因此，一個不受環境光源及背景影響的寬鬆手勢允許小幅度上下左右轉動的辨識系統是被需要的。本論文的研究目的是建立一套以深度資訊為基礎的寬鬆靜態手勢辨識系統，可以辨識十五種常用手勢。我們所提出的辨識系統主要分為兩個部份：手部偵測與手勢辨識。在手部偵測階段，我們從 Kinect 只擷取深度影像，解決手部偵測受到環境光源及背景影響的問題；並使用 Kinect SDK 提供的人體骨架定位手部區域。在手勢辨識中，我們先建立輪廓與掌心距離的一維函數 (signature)，開始偵測與篩選手部特徵點；並且根據手部輪廓上的特徵點以定位手指根部的掌指關節 (Metacarpophalangeal Joints, MCP) 位置；接著，計算每根手指的 MCP、掌心與判斷基準點所構成的夾角，作為手指角度特徵；並建立五根手指角度差異表作為手勢辨識的基準；最後使用手指數量與手指識別角度來辨識寬鬆靜態手勢。本研究實驗展示，系統整體手指數量辨識正確率有 95% 以上，在辨識標準手勢的正確率能達到 90.03%，而辨識寬鬆手勢正確率也達到 80.24%，其辨識寬鬆手勢使用多邊形逼近的輪廓優於原始鋸齒狀的輪廓。 In recent year, gesture recognition is an important issue in the field of human computer interaction. The most commonly used applications include game control, home appliances control, robot control, etc. Moreover, due to the effect of lighting, complex backgrounds and the restricted standard gestures without curvature of the fingers, some gesture recognition methods systems are neither intuitive nor comfortable for users. Thus, a loose gesture recognition system against lighting and complex backgrounds is needed. The purpose of this thesis is to develop a depth-based loose static gesture recognition system which could recognize fifteen common gestures. The proposed gesture recognition system includes two parts: hand detection and gesture recognition. In hand detection, we only capture the depth map against lighting and complex backgrounds from Kinect, and then we locate the hand region from the depth map with skeleton tracking using Kinect SDK. In gesture recognition, first, we create a signature which is a 1-D functional representation of the hand boundary. It is formed by plotting the distance from the center of palm to the hand boundary. Second, we detect features in the signature, and then we locate the Metacarpophalangeal joints of each finger from the hand region. Third, we calculate the angle of each finger by specifying three points: the center of palm at the vertex and then the anchor point and the Metacarpophalangeal joint on the rays. Finally, the system will identify each finger based on an angle table, using the number of fingers and their angles to determine the loose static gesture. In this thesis, experiments show the accuracy is up to 95% in finger counting. In recognition accuracy, 90.03% for standard gestures, 80.24% for loose gestures, and hand shape using polygonal approximation is better than original hand shape.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	740	View/Open

社群 sharing

Loading...