Research On Smoke Detection Method For High Definition Video Based On Spatial-and Frequency-domain Fusion Features

Fire is one of the major disasters that seriously endanger the safety of human life and property.Timely warning of fires is of great significance for reducing property losses.Generally speaking,smoke is often a precursor to fires,therefore,an effective smoke detection method that enables the fire to be quickly detected and controlled can effectively avoid the serious consequences of fire.Conventional fire smoke detection usually uses sensors such as light,smoke,and temperature sensors.However,these sensors have the drawbacks that only specific types of smoke can be detected and their use is limited in open space environments.Given the popularity of electronic cameras,a large number of video surveillance devices have been installed in many public places and buildings.Video surveillance-based systems provide effective coverage for larger areas and can be easily integrated into existing closed-circuit monitoring systems,smoke detection using video surveillance has become an important research topic.With the popularity of high-definition cameras and videos,the performance of an original method is often not satisfactory.This paper focuses on the smoke detection methods of high-definition video.The main research work is as follows:First,this paper studies the feature extraction and feature fusion methods of smoke images.Smoke has the characteristics of not fixed contours and translucency,so smoke and other objects are mainly developed around local features in the image.In terms of spatial domain feature extraction of smoke images,in view of the shortcomings of local binary pattern and central symmetrical local binary pattern,this paper proposes central symmetric gradient compensation-local binary pattern.In terms of feature extraction in frequency domain,a local phase quantization algorithm is proposed.This paper proposes a local phase quantization algorithm based on VLAD.After the features of the image are extracted,feature fusion is needed.This paper proposes a trisection feature fusion method.Then,this paper studies the extraction method of suspicious smoke area.Research and comparison of common motion region extraction methods,in view of the "ghost" phenomenon of ViBe algorithm,an improved ViBe algorithm is proposed as a method for extracting suspicious smoke regions.Finally,based on the above research,this paper proposes a smoke detection framework based on multi-domain fusion features for high-definition video.The framework is divided into an offline training phase and a video detection phase.This paper explains the software design and implementation of a video smoke detection system.The experiments show that the framework and algorithm proposed in this paper have strong robustness in the detection of video smoke and can identify the smoke in the video well.