Research On Fire Recognition Algorithm Based On Video Surveillance

Natural disasters are the public enemies of mankind.In many cases,human beings are powerless in the face of natural disasters.However,with the development of science and technology,there is a natural disaster that can be detected and prevented in advance.It is fire.Fires usually occur over a long period of time,and before then,if humans can detect them early enough,they can prevent them from happening.The fire was preceded by the flame and smoke characteristics,which are the two key elements of fire feature identification.In recent years,with the rapid development of video image and computer technology,video fire detection has replaced the traditional weak sensor,and become the most important fire detection method.At present,video detection technology has become more mature,but the accuracy of motion feature extraction and recognition in video images needs to be improved.In addition to the traditional moving target detection and fire feature recognition,multidimensional feature recognition and fusion techniques are needed to improve the real-time and accuracy of video detection.This paper mainly studies the fire recognition algorithm based on video.It can be divided into three sub-algorithms,moving object extraction algorithm,eigenvector extraction algorithm,improved particle swarm optimization algorithm support vector machine classification algorithm.The three algorithms realized different functions and constitute a fire recognition algorithm based on video surveillance.The main contents of this paper are as follows in fire video detection: 1.Extraction of moving objects in video images.Moving target extraction is the first and most important step of fire video recognition.The precision and timeliness of moving target extraction will directly affect the processing effect of the whole video.In this paper,we first study the basic frame difference method,optical flow method and background difference method,and analyze the merits and demerits of various algorithms for moving objects extraction.Then,the Gaussian mixture background model and ViBe background model are studied.Based on their respective models,the advantages and disadvantages of the two methods for moving target extraction are compared.The Gaussian mixture background model is finally determined as the algorithm for moving target extraction in this paper.2.A flame and smoke feature recognition and fusion algorithm is proposed.Most of the existing algorithms target smoke or flame,but the effect of single feature recognition is not perfect.On the basis of studying a large number of algorithms for the extraction and recognition of flame and smoke features,this paper compares the features in the flame and smoke,and selects the common features of the flame and smoke fire as the common features to identify the confluence of the two kinds of fire,It includes color features,area features,texture features and contour features.The four common features of smoke and flame are fused into one feature vector,which is used as the common parameters for training model to identify smoke and flame characteristics during fire occurrence.3.A support vector machine(SVM)classification algorithm based on particle swarm optimization(PSO)is proposed.Support Vector Machine(SVM)model based on fire video images containing flame and smoke to identify flame and smoke features in video images more clearly and accurately,In this chapter,PSO algorithm with mutation operation and nonlinear dynamic adjustment of inertia weights is proposed to optimize two parameters of SVM.A lot of experiments show that this algorithm has better effect than the traditional algorithm.4.Design and implementation of fire detection algorithm.Combining moving object extraction,feature vector extraction and support vector machine(SVM)classification algorithm based on Improved Particle Swarm Optimization(PSO),a complete fire identification algorithm based on video surveillance is proposed.A large number of experimental results show that the proposed algorithm has good timeliness and accuracy in fire identification.Two groups of comparative experiments have been done,and two different classifier parameters optimization algorithms such as grid search,genetic algorithm,particle swarm optimization and Bayesian classifier,BP neural network are compared.From the statistical results of two groups of experiments,it is found that the proposed algorithm has better effect than the traditional method.In summary,the algorithm in this paper has better performance in both accuracy and timeliness,which verifies the effectiveness of the algorithm.