基于视频内容分析的火灾烟雾检测算法研究与实现

发布时间：2018-04-04 02:41

本文选题：图像分类　切入点：火灾烟雾检测　出处：《北京交通大学》2017年硕士论文

【摘要】：火灾是自然人为双重诱因导致的最常见的灾难性事件之一,对人们的生产生活安全都造成了极大的威胁。而烟雾作为火灾表现出的最为显著的特性之一,准确而实时地对各种环境下的监控视频进行烟雾检测,是减小火灾危害的重要手段。近年来,随着计算机视觉和深度学习等领域的飞速发展,基于视频图像内容分析的火灾烟雾检测算法逐渐代替了传统的物理烟雾感知器。目前在机器学习领域,车辆识别、人脸识别等技术已愈加成熟,但是对于视频图像中烟雾的检测,很多图像分类方法都难以保证检测结果的准确性和检测过程的实时性。因此,本文提出了以下两种视频图像的烟雾检测算法:(1)提出了基于特征融合的SVM分类视频烟雾检测算法。建立烟雾图像标准数据库,提取正样本烟雾图像的颜色特征和纹理特征,经过特征融合作为训练输入,最终得到具有烟雾检测功能的分类器。并提出在对视频图像进行烟雾检测时,先对视频帧进行基于混合高斯模型的运动区域提取,以提高算法准确率。(2)提出了基于深度卷积神经网络的视频烟雾检测算法。设计了一种深度卷积神经网络模型,给出了基本网络结构的设置。基于本文建立的烟雾图像标准数据库对实验数据集进行了选取,并通过训练得到一个具有烟雾检测功能的CNN模型。测试实验结果显示出了较高的分类检测准确率。选取现有的经典深度卷积网络模型进行了对比实验,对实验结果的分析表明,本文所提出的网络模型对于视频烟雾检测场景具有良好的准确性和实时性。
[Abstract]:Fire is one of the most common calamitous events caused by both natural and man-made causes, and it poses a great threat to the safety of people's production and life.Smoke, as one of the most significant characteristics of fire, is an important means to reduce fire hazard by accurate and real-time detection of smoke from monitoring video in various environments.In recent years, with the rapid development of computer vision and depth learning, the fire smoke detection algorithm based on video image content analysis has gradually replaced the traditional physical smoke sensor.At present, in the field of machine learning, vehicle recognition, face recognition and other technologies have become increasingly mature, but for the detection of smoke in video images, many image classification methods are difficult to ensure the accuracy of detection results and real-time detection process.Therefore, in this paper, we propose the following two video image smoke detection algorithms: 1. We propose a SVM classification video smoke detection algorithm based on feature fusion.The standard database of smoke images is established to extract the color features and texture features of the smoke images of positive samples. After feature fusion as the training input, a classifier with smoke detection function is obtained.A video smoke detection algorithm based on deep convolution neural network is proposed to extract the moving region of video frame based on mixed Gao Si model in order to improve the accuracy of the algorithm.A deep convolution neural network model is designed and the basic network structure is set up.The experimental data set is selected based on the smoke image standard database established in this paper, and a CNN model with smoke detection function is obtained by training.The experimental results show that the accuracy of classification detection is high.The experimental results show that the proposed network model is accurate and real-time for video smoke detection.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X932;TP391.41

【参考文献】