手勢辨識(Gesture Recognition)是目前熱門研究的主題，應用的範圍非常廣，包括機器人操控，遊戲，或是家電的控制等，本論文目的在設計一個操作機上盒的人機介面，以一台攝影機即時擷取手勢影像，來達成即時手勢辨識目的，系統主要以CamShift方法動態追蹤手部並透過數位影像處理中的背景相減、形態學的運算及邊界偵測取出手部輪廓影像，而後依照手指指尖特徵找出手指並計算手指與手掌重心的角度，利用手指的數目與角度關係來判斷手勢指令，最後再透過紅外線發射裝置將指令送至機上盒。 結果驗證部份，我們定義八種手勢，分別由七位測試者針對五種狀況測試，整體平均辨識率為94.1%。 Gesture recognition is a popular research topic at present, its application is broad which includes robot operation, controller for video games or other household appliances and etc. The purpose of this paper is to design a Human-Machine interface by using a video camera to capture live hand gesture to achieve the goal of hand gesture recognition. This system will track the hand movement using background subtraction, morphology and CamShift algorithm to extract the hand contour. After the hand contour is obtained, the system will then identify each finger according to its unique feature and calculate the angle between fingers and mass centre of palm, using the number of fingers and its angles to determine the command given by the gesture. Finally, the command will be passed to STB through an infrared transmitter. For result verification we have defined 8 types of hand gesture, the experiment is conducted with 7 participants aiming at 5 scenarios. The average recognition rate is 94.1%.
