Research On Fire Detection Method For Large Space Buildings Based On Video Keyframe Extraction

Fire is an extremely dangerous emergency,which is very destructive.If the fire is not controlled,it will lead to huge disasters and immeasurable losses to human,economy and ecology.The role of large space buildings disaster rescue in urban safety management has attracted more and more attention.As one of its core components,the importance of fire detection for large space buildings is also self-evident.For the fire of large space buildings,research on fire detection method based on video keyframe extraction is conducted in this paper.The SuperPoint method and the improved SuperPoint method using elastic distortion to extract video keyframe are introduced.The convolution neural network architecture and main steps of fire identification and detection method based on keyframe and superpixel are explained.Finally,for the engineering requirements of large space buildings,a fire detection system for large space buildings is designed and implemented.The main research works of this paper are as follows:Firstly,in order to improve the speed of fire detection,a keyframe extraction method based on improved SuperPoint is designed in this paper.The keyframe extraction strategy given in this paper avoids detecting each frame of surveillance video,and only identifies and detects the fire on the extracted video keyframe,which improves the detection speed to a great extent.This paper introduces the training process of SuperPoint method in detail,and explains the principle of using elastic distortion to improve SuperPoint method,and expounds the main steps of using improved SuperPoint method to extract keyframe.The experimental results show that this method shows good performance in interest point extraction,descriptor calculation,interest point detection and matching,and realizes good result of video keyframe extraction,which can extract more realistic video keyframe.Secondly,aiming at the problems of slow detection speed,low identification accuracy and low detection precision of fire intelligent surveillance system,a fire identification and detection method based on keyframe and superpixel is designed in this paper,which makes full use of the characteristics and advantages of deep learning method.The network architecture and main steps of fire identification based on convolutional neural network and fire detection based on superpixel and convolutional neural network are introduced in detail.The experimental results show that this method shows good performance in fire identification and detection of different video keyframe and large space buildings,and realizes good results of fire identification and fire detection,which can accurately identify whether there is a fire and precisely detect the location of the fire.Finally,for the engineering requirements of large space buildings,combined with the keyframe extraction method and the fire identification and detection method designed in this paper,this paper designs and implements a fire detection system for large space buildings.The system mainly includes the methods of keyframe extraction based on improved SuperPoint,fire identification based on convolutional neural network and fire detection based on superpixel and convolutional neural network.It realizes three main functions: video show,video start and stop and video reset.The experiments show that the system can quickly and accurately identify whether there is a fire in the surveillance video and precisely detect the location of the fire,and realizes good result of fire detection for large space buildings,which meets the requirements of fire detection for large space buildings.Figure [41] Table [8] Reference [56]...