利用光流計算來作煙霧偵測

Abstract

近年來數位式錄影系統(DVR)廣被應用在各種監控場所，因為其數位化的優點，使得有相當多的影像及視訊處理技術可以被嵌入結合在一起，如移動偵測、車流分析、自動車牌辨識等等，因此提升了數位錄影系統整體的應用價值。 本篇論文主要在研究一種以RGB彩色模型為基礎的煙霧偵測方法，首先對一般煙霧的顏色特徵做彩色影像的分析，得到一組以RGB模型參數代表的模組，直接使用這個模組將煙霧從影像中分離出來，對分離後的影像依據煙霧會向上竄升的特性，利用光流法去求出煙霧內部的運動情形，配合煙霧的增長再利用模糊推論系統判斷移動物體是否為煙霧，適時的提供警報訊息。而這裡所提出的方法，和傳統的煙霧偵測器截然不同，將來有機會被應用於空曠的戶外空間，預防森林大火的發生，經由實驗的驗證，若移動物體的呈現面積夠大，可得到不錯的效果。

In recent year, digital video record system(DVR) is widely applied to various kinds of surveillance environment. Because of the advantage of digitization, a lot of image processing technology can be imbedded and combined together. In this thesis, study on a new method of smoke detection based on RGB color model mainly. At first, using a mould that represents with RGB model parameters separates out the smoke from the image and according to the characteristic that smoke moves upward. Utilize optical flow method to know the sport situation within the smoke and cooperate with the growth of smoke. Finally, fuzzy inference system judge the moving object is smoke or not. Then offer alarm information in right time. The proposed method is completely different to traditional smoke detector. This will have an opportunity to be applied on the outdoor space to prevent the emergence of the forest fire in the future. From the experiment results, if the area of appearing of the moving object is large enough, we can get satisfied result.