基於手形輪廓之智慧型手勢辨識系統設計

Abstract

本論文主要是研究可應用於手勢辨識之智慧型辨識系統，其設計流程共分為三個部分，首先是利用影像處理取得手勢的影像，並以手形輪廓掃描的方式描述手勢的特徵，其次是採用類神經網路的架構，以倒傳遞法則來訓練類神經網路的參數，使其得以辨識手勢的特徵，最後再利用實驗來驗證將倒傳遞類神經網路應用於手勢辨識的可行性，由實驗結果可知，本論文所開發之智慧型辨識系統可以完全準確地辨識出數字1至5的單純手勢，而對於較為複雜的數字0至9，也可利用提高手勢特徵的精準度，以及增加類神經元的方式，達到準確辨識目的。

This thesis proposes an intelligent recognition system based on neural network for hand gesture recognition. The design process is partitioned into three parts. First, the hand gesture is abstracted from an image by image procession and the characteristics of hand gesture are retrieved by the hand contour scanning method. Then, based on the back-propagation algorithm, the neural network is trained to learn the charateristics of the hand gesture. Finally, experiments are included to demonstrate the feasibility of the developed intelligent hand gesture recognition system. From the experimental results, it is clear that the system can precisely recognize the simple hand gestures related to the numbers from 1 to 5. For the more complicated case to expresss 0 to 9, the system is also available to successfully recognize all of them, after increasing the resolution of hand gesture and the number of neurons in the neural network.