台灣人口主要集中在西部平原的六都地區，而人口密集伴隨的是火災案件頻傳，其中許多火災是由於老舊建築缺乏完善的消防設備所致。儘管在一般住宅中普遍安裝了煙霧和溫度感應器，但由於安裝位置的限制，有時無法及時發出警報。故本研究利用攝影機之影像辨識的方式，讓火源在初期即可發現並予以撲滅。
考慮到室內攝影機的不同格局和安裝位置可能存在死角，因此採取了在天花板和地板分別安裝兩個攝影機進行觀測分析。通過使用Pyrosim軟體建立空間模型後，將火源分為極快速、慢速和爐火三種模式，並在房間內不同位置放置火源進行模擬，再把分析結果之溫度分布圖利用Python程式碼轉換成灰階值圖，針對不同火源案例、不同時間點之最大灰階值總結出火災判斷基準。
最終判斷基準如下:攝影機每5秒會記錄一張溫度分布圖並轉成灰階值圖，如果前後兩張的灰階值不同，將前一張圖片視為初始秒數並開始相減計算，後續如地板灰階值持續增加並超過0.4，或者天花板灰階值持超過0.2時即發出警報。若灰階值在一段時間後為一定值或減少，計算將停止。
研究結果顯示，根據地板判斷基準，警報可提前至少20秒於溫度感應器發出；而根據天花板判斷基準，警報可提前至少30秒於溫度感應器發出。

This research proposes the use of image recognition from cameras to detect and extinguish fires early.
Considering that indoor cameras may have different perspectives and blind spots due to their installation positions, two cameras are installed for observation and analysis—one on the ceiling and one on the floors. Constructing a spatial model by using Pyrosim software, categorizing fire sources into three modes: rapid, slow, and furnace fire. Then simulated the fire incidents by placing fire sources at different positions within the room. The temperature distribution maps resulting from the analysis are converted into grayscale images using Python code. Based on different fire scenarios and time points, the maximum grayscale values are summarized to establish fire detection criteria.
The final criteria as below: The camera records a temperature distribution map and converts it into a grayscale image every 5 seconds. If the grayscale values of consecutive images differ, the previous image is considered the initial time, and subtraction calculation would begin. Subsequently, if the grayscale values on the floor continuously increase and exceed 0.4, or if the grayscale values on the ceiling exceed 0.2, an alarm is triggered. If the grayscale values remain constant or decrease after a certain period, the calculation would stop.
The research results indicate that based on the floor detection criteria, the alarm can be triggered at least 20 seconds earlier than temperature sensors, while based on the ceiling detection criteria, the alarm can be triggered at least 30 seconds earlier than temperature sensors.

[1] 沈鈺棠，「集合住宅火災模擬分析-以台北縣新莊鈴木華城集合住宅為例」，國立雲林科技大學環境與安全衛生工程研究所，2009
[2] 王思文，「老人住宅之規劃設計與火災逃生動態模擬」，國立成功大學土木工程研究所，2015
[3] 嚴潤民，「應用FDS模擬歷史性建築火場之研究-以原台南愛國婦人會館為例」，國立成功大學建築研究所，2015
[4] 郭榮宏，「運用火災調查技術與FDS電腦模擬程式重建桃園新屋保齡球館火災情境」，吳鳳科技大學消防系碩士在職專班，2017
[5] 潘文貞，「FDS運用在透天住宅火災減災之探討-以高雄地區為例」，國立高雄科技大學土木工程系碩士在職專班，2021
[6] 陳峰彬，「運用FDS模擬醫院火場行為」，國立成功大學土木工程研究所，2022
[7] 蔡宜霖，「火災模擬與實尺寸燃燒驗證之比較研究-以原台南愛國婦人會館為例」，國立成功大學建築研究所，2014
[8] 呂淑然、楊凱，「火災逃生模擬仿真-Pyrosim+Pathfinder 中文教程與工程應用」，北京化學工程出版社，2014
[9] 劉嘉修護理之家，精緻單人套房，查詢日期2024年5月06日
[10] 「火災案件搶救出勤紀錄表填寫作業原則」，內政部消防署，2016
[11] 「火災災害防救業務計畫」，內政部，2019
[12] 「火災統計表」，內政部消防署，查詢日期2024年3月26日
[13] 大翰科技股份有限公司，熱點及火焰偵測智慧攝影機產品介紹，查詢日期2024年3月28日
[14] 柏聯科技，影像管理軟體-AI火焰與煙霧偵測 ，查詢日期2024年3月28日
[15] Kevin McGrattan, Simo Hostikka, Jason Floyd, Randall McDermott, Marcos Vanella ,Eric Mueller,“Fire Dynamics Simulator User’s Guide”, National Institute of Standards and Technology, 2013
[16] Liang-Sheng Wu , Ryh-Nan Pan , Shiuan-Cheng Wang , Chung-Hwei Su , Wen-Chien Wu, “Characteristic Analysis of Four Major Nighttime Fire Cases on Fire Safety of Long-Term Care Institutions Using Fire Protection Defense-In-Depth Strategy”, Fire , 2023
[17] Youcun Lu , Lin Duanmu , Zhiqiang (John) Zhai , Zongshan Wang, “Application and improvement of Canny edge-detection algorithm for exterior wall hollowing detection using infrared thermal images”, Building and Environment 274, 2022
[18] Chenqiu Du , Yujue Wang , Baizhan Li , Mengyu Xu , Sasan Sadrizadeh, “Grey image recognition-based mold growth assessment on the surface of typical building materials responding to dynamic thermal conditions”, Building and Environment Vol. 243, 2023
[19] Tianhang Zhang , Fangqiang Ding , Fu Xiao , Chris Xiaoxuan Lu , Xinyan Huang, “Forecasting backdraft with multimodal method: Fusion of fire image and sensor data”, Engineering Applications of Artificial Intelligence, Vol. 132, 2024
[20] Jewon Oh , Yutaka Matsumoto , Kohei Arai, “Development of Crack Detection and Crack Length Calculation Method using Image Processing”, International Journal of Advanced Computer Science and Applications, Vol. 14, 2023
[21] Md Sultan Mahmud , Long He , Azlan Zahid , Paul Heinemann , Daeun Choi , Grzegorz Krawczyk , Heping Zhu, “Detection and infected area segmentation of apple fire blight using image processing and deep transfer learning for site-specific management”, Computers and Electronics in Agriculture, Vol. 209, 2023
[22] Tami M. Brown-Brandl , Roger A. Eigenberg , Joseph L. Purswell, “Using thermal imaging as a method of investigating thermal thresholds in finishing pigs”, Biosystems Engineering,Vol. 114 , 2013
[23] Parkhide Hassani , Fatemeh-Sadat Izadi-Avanji , Mahnaz Rakhshan , Hamid Alavi Majd, “A phenomenological study on resilience of the elderly suffering from chronic disease: a qualitative study”, Dove press open access to scientific and medical research, 2017
[24] J. Glasa, L. Valásek, P. Weisenpacher, L. Halada, “Cinema Fire Modelling by FDS”, Slovak Academy of Sciences/Institute of Informatics, 2013
[25] Mingbiao Xu , Dehong Peng, “PyroSim-Based Numerical Simulation of Fire Safety and Evacuation Behaviour of College Buildings”, School of Construction Engineering, Yiwu Industrial & Commercial College, 2020
[26] Tianhang Zhang , Fangqiang Ding , Zilong Wang , Fu Xiao , Chris Xiaoxuan Lu , Xinyan Huang, “Forecasting backdraft with multimodal method: Fusion of fire image and sensor data”, Engineering Applications of Artificial Intelligence 132 , 2024
[27] Tami M. Brown-Brandl , Roger A. Eigenberg , Joseph L. Purswell , “Using thermal imaging as a method of investigating thermal thresholds in finishing pigs”, Biosystems Engineering Volume 114, Issue 3, 2013