多特征融合的火焰检测算法

2016-01-18吴茜茵,严云洋,杜静等

智能系统学报 2015年2期

网络出版地址：http://www.cnki.net/kcms/detail/23.1538.TP.20150302.1106.009.html

多特征融合的火焰检测算法

吴茜茵1,2，严云洋1,2，杜静1,2，高尚兵2，刘以安1

(1.江南大学物联网工程学院，江苏无锡 214122； 2. 淮阴工学院计算机工程学院，江苏淮安 223003)

摘要：视频火焰检测是复杂场景下预防火灾的重要方法。为了提高火焰的检测效率和鲁棒性,基于RGB和HSI颜色空间改进了火焰的颜色特征模型，有效地提取了疑似火焰区域；实验对比分析了火焰不同的形状结构特征，及其特征组合对火焰检测有效性的影响，提出了一种融合圆形度、矩形度和重心高度系数的火焰检测算法，然后将融合后的火焰特征输入支持向量机(SVM)中进行分类。在Bilkent大学火灾视频库上的实验结果表明，该方法高效、快速，且能适用于多种场景。

关键词：特征提取；特征融合；支持向量机；颜色模型；火焰检测；圆形度；矩形度；重心高度

DOI:10.3969/j.issn.1673-4785.201406022

中图分类号：TP391.41文献标志码：A

收稿日期：2014-06-13. 网络出版日期：2015-03-02.

基金项目：教育部科学技术研究重大资助项目(311024)；国家自然科学基金资助项目(61402192)；江苏省“六大人才高峰”资助项目(2013DZXX-023)；江苏省“333工程”资助项目；江苏省“青蓝工程”资助项目；淮安市“533工程”资助项目(BRA2013208)；淮安市科技计划资助项目(HAG2013057，HAG2013059).

作者简介：

中文引用格式：吴茜茵，严云洋，杜静，等. 多特征融合的火焰检测算法[J]. 智能系统学报， 2015, 10(2): 240-247.

英文引用格式：WU Xiyin， YAN Yunyang， DU Jing， et al. Fire detection based on fusion of multiple features[J]. CAAI Transactions on Intelligent Systems, 2015, 10(2):240-247.

Fire detection based on fusion of multiple features

WU Xiyin1,2， YAN Yunyang1,2， DU Jing1,2， GAO Shangbing2， LIU Yi’an1

(1. School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, China; 2. Faculty of Computer Engineering, Huaiyin Institute of Technology, Huaian 223003, China)

Abstract：Video fire detection is an important method to prevent fire disaster under complex circumstances. In order to improve the efficiency and robustness of fire detection, the color feature model can be improved based on RGB and HSI color space and the suspected flame area is extracted effectively. After analysis on the experimental results with different features of shape or structure of fire and the influence of their combined features on the validity of fire detection, a method of flame detection is proposed based on fusion of circularity, rectangularity and the coefficient of orthocenter height. Based on fusion of these flame features, the support vector machine (SVM) is used for classification. Experimental results on the fire videos at Bilkent University show that the proposed algorithm is efficient and fast for fire detection， and it could detect fire real-time under a variety of circumstances.

Keywords：feature extraction; feature fusion; support vector machine; color model; fire detection; circularity measures; rectangularity; orthocenter height