基於影像特徵點擷取結合深度資訊之 即時手勢辨識系統設計

Abstract

近年來，手勢辨識可應用的領域相當廣泛，因此受到重視且被深入的研究與探討，例如人機溝通、遠距遙控等皆是。一般而言，手勢辨識系統先根據手勢模型找出其特徵，再利用這些特徵來做辨識。本篇論文提出以Kinect之彩色及深度影像為基礎的手勢特徵擷取，來設計即時手勢辨識系統。整個系統分成三個部分：前景偵測、特徵擷取及手勢識別。首先利用膚色偵測配合聯通物件法濾掉背景並找出可能的手勢範圍；之後藉由距離轉換並尋找距離轉換區域的最大值來萃取特徵點，進而利用特徵點找出手勢特徵，包括手的方向、掌心位置、指尖位置及指向。最後，設計以手勢特徵為依據之即時手勢辨識系統，從實驗結果可知，本論文所提之手勢辨識系統確實可以達成具有成效之即時辨識功能。

In recent years, hand gestures recognition(HGR) approaches have been widely applied to a diversity of areas, like human-computer interface(HCI) and remote control systems. The HGR systems usually rely on a hand model to extract useful hand gesture features. This thesis proposes a robust and fast feature extraction method for hand gesture based on the depth and RGB information from Kinect to implement a real-time HGR system. The system is divided into three parts, including region-of-interest (ROI) selection, feature extraction and hand gesture recognition. First, the skin color detection and connected component labeling(CCL) are applied to select the potential ROIs. Then, pixels with local maximum distance-transformation-value in the potential ROIs can be extracted as feature points. Further, these feature points could be used to find hand gesture features such as hand orientation, palm center and fingertip positions and directions. Finally, the extracted hand gesture features are send into the HGR system. From the experimental results, the proposed hand gesture recognition system can perform in real-time and possess good recognition rates.