作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2018, Vol. 44 ›› Issue (7): 264-270. doi: 10.19678/j.issn.1000-3428.0048082

• 图形图像处理 • 上一篇    下一篇

基于深度森林模型的火焰检测

朱晓妤 1,2,严云洋 1,2,刘以安 1,高尚兵 2   

  1. 1.江南大学 物联网工程学院,江苏 无锡 214122; 2.淮阴工学院 计算机与软件工程学院,江苏 淮安 223003
  • 收稿日期:2017-07-24 出版日期:2018-07-15 发布日期:2018-07-15
  • 作者简介:朱晓妤(1993—),女,硕士,主研方向为数字图像处理、模式识别;严云洋(通信作者)、刘以安,教授、博士;高尚兵,副教授、博士。
  • 基金资助:

    国家自然科学基金(61402192);江苏省“六大人才高峰”项目(2013DZXX-023);江苏省“333工程”项目(BRA2013208);淮安市科技计划项目(HAG2013057,HAG2013059)。

Flame Detection Based on Deep Forest Model

ZHU Xiaoyu 1,2,YAN Yunyang 1,2,LIU Yian 1,GAO Shangbing 2   

  1. 1.School of Internet of Things Engineering,Jiangnan University,Wuxi,Jiangsu 214122,China; 2.Faculty of Computer and Software Engineering,Huaiyin Institute of Technology,Huaian,Jiangsu 223003,China
  • Received:2017-07-24 Online:2018-07-15 Published:2018-07-15

摘要:

在进行视频火焰检测时,周围环境以及火焰本身亮度的变化会对背景建模造成影响。针对该问题,提出一种基于帧频提升的高斯混合背景建模方法。在当前帧和前一帧之间插入若干帧,使高斯混合模型构建出的背景更贴近当前帧的真实背景,有利于后续的目标检测和目标提取。同时,构建一种应用于火焰检测的深度森林模型,对基于帧频提升的高斯混合背景建模方法所提取的火焰候选区域,先使用双视角、深层多粒度扫描结构提取出其抽象特征,再使用深度森林模型进行火焰检测。实验结果表明,该方法能够增强火焰特征的抽象表示能力,提高火焰检测率,并且具有强鲁棒性。

关键词: 火焰检测, 深度森林, 高斯混合模型, 帧频提升, 背景建模

Abstract:

Background modeling for video flame detection is related with the surrounding scene which is influenced by the change of surrounding environment and the brightness of the flame.Aiming at this problem,a novel method of Gauss mixture background modeling based on frame rate up-conversion is proposed.Firstly,several frames are inserted between the current frame and the previous frame,so that the background constructed by the Gauss mixture model is more similar to the real background of the current frame.It is more favorable for the subsequent target detection and target extraction.Then the deep forest model for flame detection is presented.Flame can be detected by using this deep forest model after the abstraction features of candidate regions being extracted by the dual vision and deep multi granularity scanning structure.Experimental results show that the proposed method improves the flame detection rate and has better robustness because the representation ability of flame characteristics is enhanced.

Key words: flame detection, deep forest, Gauss mixture model, frame rate up-conversion, background modeling

中图分类号: