Research On Video-based Smoke Detection Algorithms

Smoke is one of the most prominent features of the early fire.Accurate detection of video smoke in various environments is an important means of reducing fires and is of great significance to society With,the development of artificial intelligence technology,research on smoke detection technology based on computer vision has received extensive attention.This paper studies the smoke detection algorithm based on computer vision.The mail work of this thesis is as follows:(1)In the candidate region extraction phase,three common motion zone extraction algorithms are compared and analyzed,such as optical flow method,background difference method and frame difference method.Finally,we select the background difference method for smoke candidate zone extraction,combined with dark channel prior.(2)The smoke detection method based on wavelet transform is studied.The multi-level wavelet deconposition is combined withat th LBP and HOG features to extract smoke features.The experimental results show the method is accurate for video smoke detectio.The rate is higher than the existing method and reduces the false detection rate.(3)Studying the application of the LGBP method to smoke detection.Applying LGBP to smoke detection can further extract multi-directional and multi-scale local binary information of smoke images,and enhance the desoription of features.Experiment shows that the accuracy of video smoke detection based on LGBP method is higher than the common feature extraction methods,and the alse detection rate is greatly reduced.(4)This thesis explores a method of combining convolutional neural network and SVM to detect smoke in video,discusses a SqueezeNet-SVM network model,and compares it with common deep learning networks.Our network accuracy is higher than that common deep learning networks and reduced the rate of false positives.