Video Smoke Detection Based On Convolutional Neural Network

Once fire occurs,it will cause immeasurable damage.As soon as the fire is discovered,it can be extinguished as soon as possible,and the loss will be minimized.When fire occurs,smoke usually generated first,so the detection of smoke can detect the danger earlier.With the development of technology,video smoke detection technology appears.This is a method of using computer vision technology for smoke detection.It has the characteristics of low cost,short response time and large coverage area.Most of the existing video smoke detection technologies use one or more characteristics of smoke.However,the characteristics of smoke are highly susceptible to the surrounding environment and the combustion products,use these methods is prone to false positives and false negatives.In recent years,the technology of deep learning has developed rapidly and has penetrated into various fields.Convolutional neural networks are an important branch of deep learning,it has powerful image recognition capabilities.The deep convolutional neural network can learn the features of the bottom to the top of the image from a large amount of training data,use these features can better express the essence of things.Compared with other smoke detection methods,the convolutional neural network which trained by a large number of smoke data can learn the complex features of smoke in different scenes,and can detect smoke more accurately.Therefore,in this paper proposes a video smoke detection method based on convolutional neural network.First,using variety of training data to train the convolutional neural network to identify multiple moving targets including smoke.After this,preprocess the video,since the smoke has a floating motion characteristic,the static interference in the video can be removed by the mo ving target detection,most of the non-smoke regions are filtered out,then the moving target image extracted by the video pre-processing is sent to the convolutional neural network for identification and keep the images recognized as smoke,then mark these areas in the video with a red box to complete the smoke detection.Video smoke detection system is completed with Python,TensorFlow,Opencv.The detection process uses GPU to speed up processing.After properly adjusting the video resolution,it can meet the requirements of real-time detection,and four videos can be processed simultaneously with multi-threading.With the powerful image recognition capability of the convolutional neural network,the detection accuracy is further improved compared with the traditional smoke detection method,and false positive rate and false negative rate are greatly reduced.