Research On Smoke Video Image Recognition Algorithms

With the rapid development of the society and economy,fires occur frequently on large space,which has created a significant threat to human lives and property,but conventional fire detectors are affected by various factors such as space height,dust etc,so it can't play a proper role.With the development of information technology,such as image processing,pattern recognition,computer vision,image fire detection technology is one of the most important research for the early fire detection,when fire will occur,the smoke is early in the flame,there smoke detection becomes a top priority of fire detection research.This article combines the static texture features with the dynamic movement of smoke to detect and recognize for video image,find out the fire information,realize early warning of fire in large space environment.At first,this paper analyses advantages and disadvantages of some moving targets detection algorithm,then propose the color model combined with background subtraction algorithm to extract the suspected smoke area in the video image;Followed by extracting the static texture features of smoke,using the local binary pattern based on image pyramid and local binary pattern variance based on image pyramid algorithm to describe the texture information of smoke,which can get the local and global texture characteristics of smoke,through experiment verify that this method is a good way to represent the texture feature of smoke;And then the image block processing technique combined with improved Lucas&Kanada optical flow algorithm is used to extract smoke movement characteristics,which get high accuracy,reduce the complexity of the algorithm and advance the operation time;Finally extracted the multiple characteristics put in SVM as a criterion to identify the smoke whether the characteristics meet the feature criteria of smoke,if meeting the feature criteria sends out alarm signals,if not,don't to deal with,after compared with other algorithms,found that the method improves the accuracy of video smoke detection.A lot of video experiments results show that this smoke video image recognition algorithm is stable and reliable,and the algorithm achieves the desired effect with less false alarm and missed alarm can satisfy for the real-time and accuracy requirements image fire detection in the large space.