Research On Smoke Detection Technology Based On Video Image

Fire has always been a problem perplexing human society,bringing heavy economic and personal disasters to human society.The struggle between man and fire has lasted for thousands of years.In recent years,Researchers of computational science also apply the latest image processing technology to smoke detection,and develop a series of practical algorithms.As a branch of image processing,image recognition has more and more extensive application fields.This paper analyzes the current research status of video image smoke detection at home and abroad,and locates the innovation point of this paper.The three-frame difference method is combined with the Gaussian mixture background modeling method to determine the video smoke candidate area.By extracting the texture features and HOG features of the suspected smoke area as the input of the support vector machine SVM classifier,the video smoke image classification results are obtained.At the same time,the VGG16 neural network is compared with the image smoke detection effect,and an efficient smoke recognition algorithm is obtained.Finally,an automatic smoke detection system was built to realize the function of real-time video monitoring of fire and smoke.This paper mainly studies the theory and application of image recognition in the field of smoke detection.Through manual collection,artificial generation and simulation synthesis methods,the smoke video image data set is obtained.At the same time,relevant image enhancement methods are used to improve the usability of data set experiment.This paper proposes a method combining three frame difference method and Gaussian mixture background modeling to detect the moving smoke area in video.According to the smoke features in video image,the classical feature extraction algorithm is studied,and then the classic classifier SVM is used to classify the smoke features to determine whether there is smoke in video image.With the rise of deep learning technology in the world,more and more researchers begin to try the performance of neural network in the original field.This paper studies and analyzes the classification algorithm tool of deep learning,the mathematical principle of convolution neural network,and proposes u-net convolution neural network to do image segmentation instead of background subtraction to extract video motion area to improve the smoke area in video image Domain detection range.Vgg 16 neural network is proposed to train the video image to improve the low detection rate of traditional video image smoke detection technology.Finally,the excellent model of vgg16 training is applied to the smoke detection model,the automatic smoke detection system is developed,and the smoke detection model is embedded to improve the system performance.