基於凹槽匹配法之即時手勢辨識系統設計

Abstract

論文主要以模仿人眼視覺運作模式的特點來發展即時手勢辨識系統，辨識以二種手勢為主，其中一種用來控制滑鼠游標方向、另一種則用來下達點選的命令。此即時手勢辨識系統分四個階段進行，分別是移動物體追蹤、目標影像前處理、手型模板影像凹槽匹配與手勢辨識。影像中可能是手部的區域主要是以移動物體追蹤，定位出包含手部區域之區域影像之後，再利用目標影像前處理後所產生之距離圖像(distance map)，配合手型模板邊界影像凹槽匹配之後，而能在整張影像中搜尋出手部區域，並得到主要目標影像(Region of Interest, ROI)，進入第四階段後，手勢辨識系統再根據所決定之手部區域的距離圖像來進行辨識而能達到即時手勢辨識的功能。最後，從本論文之實驗結果可證明所提出的即時手勢辨識系統及其功能應用在人機界面上之可行性。

This thesis proposes a real-time hand gesture recognition system to mimic the characteristics of human vision. There are two hand gestures to be recognized, one for cursor control and the other for click control, and the system is implemented in four stages, including moving object tracking, object image preprocessing, hand template chamfer matching, and hand gesture recognition. Some regions with moving objects are determined by the moving object tracking, and then from object image preprocessing the potential hand region is detected and transformed into a distance map. Based on the distance map, the hand template chamfer matching is adopted to determine whether it is a hand or not. After the hand region is confirmed, the region of interest (ROI) is generated to contain the area that the hand may move around. In the fourth stage, the hand gesture recognition applies to judge the current gesture of the moving hand according to its distance map. Finally, experiment results are used to demonstrate the feasibility of the proposed hand gesture recognition system, which shows a way of real-time human-computer interaction.