Research Of Fire Smoke Detection Algorithm Based On Video

Fire is one of fatal disasters which endanger human life and property security and natural ecological environment seriously.Timely warning of fire is ofgreat significance to reduce the losses.The flame is small in the early of fire in general,but the smoke is very obvious,so the detection of smoke is an important basis to timely determine whether the fire occurred.The traditional fire detection techniques rely on sensors,which is difficult to implement in open spaces.The fire smoke detection technology based on video image has attracted much attention with the wide use of intelligent monitoring equipment.It can effectively avoid the influence of some environmental factors and has obvious advantages in large-scale space monitoring.This paper proposes a fire smoke detection algorithm based on video image,which combines dynamic detection and static classification.Methods such as candidate smoke regions extraction and image feature extraction and classification are studied.The main works of this paper are as follows:(1)We propose a fire smoke detection algorithm based on background dynamic updating and dark channel priori.Firstly,the improved background dynamic updating algorithm is used to extract the motion foreground,which solves the problem of the hollowness for the slowly diffusing smoke caused by the traditional motion detection algorithm.Then,aiming at the problem that the current algorithm is not adaptable in the complicated environment.For example,there are interferences such as shaking branches,pedestrians and other moving objects in the natural scenes,which prone to false alarm.We propose interference objects filtering method based on dark channel prior knowledge,the method combined with the moving target detection algorithm can eliminate the majority of interfering objects in the stage ofcandidate smoke regions extraction.Finally,the classification and identification is achievedby multi-feature fusion.Experimental results show that the algorithm can effectively reduce the false alarm and improve the performance of detection.(2)We propose a fire smoke detection algorithm based on convolutional neural networks.Because smoke has no fixed color and contour,the traditional smoke detection algorithm based on the feature of manual design is difficult to describe the essential properties of smoke,which affects the accuracy of detection.At the same time,manual design and processing of the features is need for a certain degree of expertise and experience,which bring difficulty to the research of fire smoke detection.Therefore,a fire smoke detection algorithm based on convolution neural networks is proposed on the basis of the previous research thatcan automatically learns the most discriminative high-levelfeatures through the multilayer structure of the network.The high-level features of the learning makes our algorithm is robust to significant appearance changes which is suitable for the extraction of features of changes objects like smoke.Experiments show that the strategy of implicit enlarges the candidate smoke region combined with high-capacity convolutional neural networks,which greatly improves the accuracy and timeliness of video smoke detection.