煙霧偵測上的時空分析

Abstract

近年來，基於影像式的煙霧偵測技術在智慧型監控系統中受到廣泛的重視與研究。然而，在給定一個廣大的開放空間中來處理煙霧事件與其他的常見的干擾物例如行人和車輛，建立一個穩定且有效率的煙霧偵測系統仍難是一個困難且具有挑戰性的問題。在本篇論文中，我們提出了一個創新與可靠的自動化煙霧偵測架構。本篇論文提出三種重要的特徵：邊緣模糊化、能量的逐步變化與色彩結構的逐步變化。接下來，在考量實際火災與煙霧事件的影片與資料的稀少性下，為了獲得更佳的一般性，我們採用基於支持向量機(Support Vector Machines)的分類器將三種特徵結合。此系統在各種環境與干擾下執行超過六小時以上，並且證明在實際防災應用上的穩健性與可靠性。 本篇論文的目標是在時間域與空間域上分析煙霧的特性，由此系統所得到的實驗結果將可以提供煙霧偵測領域上更深入的理解，並且有助於處理高誤判率和較長的反應時間等問題。

Visual-based smoke detection techniques in surveillance systems have been studied for years. However, given an image in open or large spaces with typical smoke and disturbances of commonly moving objects such as pedestrians or vehicles, robust and efficient smoke detection is still a challenging problem. In this paper, we present a novel and reliable framework for automatic smoke detection. Automatic monitoring systems use a stochastic approximation procedure, which is used to recursively estimate the parameters of the Gaussian mixture model and construct a background image for foreground segmentation. Next, spatial and temporal characteristics are analyzed of the candidate regions in the video sequences. It exploits three features: edge blurring, the gradual change of energy and the gradual change of chromatic configuration. In order to gain proper generalization ability with respect to sparse training samples, the three features are combined using a support vector machine based classifier. This system has been run more than 6 hours in various conditions to verify the reliability of fire safety in the real world. The results obtained from this novel approach would provide better insight to operators in the field of smoke detection to handle the problems of high false alarm rate and long reaction time.