基於區域與全域分析之影像式火焰偵測系統

Abstract

近年來，基於影像式的火焰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，在影像式偵測系統中有許多事件的干擾造成系統判斷上的困難，例如閃爍的紅光或紅色移動物體，建立一個穩定且有效率的火焰偵測系統能仍然是一個有趣且困難的挑戰。本篇論文中利用了區塊式的特徵擷取方式，可利於獲得區域內資訊並減少運算資料量。火焰的區域特徵模擬火焰物體的細部特性，分別為火焰顏色分析、火源不動性、混亂分析，三種特徵都有足夠的偵測率且可濾除不同的誤報。配合全域特徵分析火焰的整體特性，如質地分析、火焰面積分析，兩項特徵可幫助通過區域分析的鄰近候選區塊進一步的驗證，使得誤報率降低至理想的狀況。實驗結果顯示本篇論文提出的火焰偵測系統在各種不同的環境條件下擁有良好偵測率以及低誤報率，證明在實際防災應用上的穩健性及可靠性。整個系統所使用的演算法採取運算量低的計算方式，使得此系統可應用在更多的硬體設備上，提高此系統在智慧型監控系統被廣泛使用的可能性。

In recent years, visual-based fire detection technology in intelligent surveillance systems has received wide attention and research. However, many disturbances can cause problems on a visual-based detection system, such as twinkling light or red moving objects. Establishing a stable and effective fire detection system is still a difficult challenge. This thesis uses the block-based feature extraction method, which easily analyzes local information in the region and reduces the computing data. Local feature of fire block are extracted from the detailed characteristics of fire objects, which are fire color analysis, fire source immobility, and disorder analysis. All three features have great detection rate and filter out different false-positive cases. Analyzing global features with textures and fire area of local candidates selected by local feature analysis further reduces false alarms of the proposed system. Experimental results show that the fire detection system proposed in this thesis has high detection rate and low false alarm rate under various environments, and prove the reliability and stability in real fire-safety applications. The proposed system is composed of algorithms with low computation, so the system has high possibility for wide use by intelligent monitoring applications based on embedded devices.