Video Smoke Detection Based On 3D Residual Dense Network

Video smoke detection technology has the advantages of fast response,not easily Video smoke detection technology has the advantages of fast response,less environmental impact,wide application range and low cost,which provides effective protection for early fire warning.The traditional video smoke detection method mainly uses the machine learning method to classify and identify the smoke by extracting the image features of the smoke.Although the detection accuracy is improved,there are still problems such as high false positive rate and high false negative rate.Using convolutional neural network to identify images can automatically learn features from image data,better characterize the essential information of images,and facilitate classification and recognition.Aiming at the problem that can only extract the spatial information of the image of the traditional detection method and the two-dimensional convolutional neural network,these methods lead to false positives and false negatives,thesis proposes a method based on 3D convolutional neural network to detect video smoke by 3D Residual Dense Network.The method of detecting video smoke based on 3D Residual Dense Network mainly deals with video images of continuous frames.This thesis comprehensively uses block motion detection and smoke motion direction feature detection suspected smoke method to quickly filter out most non-smoke areas and reduce The time complexity of smoke detection improves the efficiency of subsequent detection.At the same time,the residual module Residual Block and Dense Block are integrated to form the Residual Dense Block(RDB),which is extended to the 3D Residual Dense Block to extract the spatiotemporal characteristics of smoke and use the public data set for Verification of algorithm performance,and comparied with current mainstream detection methods.The experiment proves that this method is obviously improved in the detection accuracy,and the false detection rate and the missed detection rate are also reduced.