Research On Indoor Video Smoke Early Warning Algorithm Based On Motion Context Information Accumulation

Fires often cause significant economic losses and casualties,especially indoor fires.Due to the narrow indoor space and the presence of other combustible materials,the escape of personnel requires a certain amount of time and other factors,increasing the probability of casualties.Therefore,the establishment of a complete indoor fire warning system is one of the important measures to prevent fires.Smoke is one of the characteristics of the initial fire.With the development of computer vision technology,intelligent video surveillance is widely used in the field of security.This paper proposes an indoor smoke early warning algorithm based on video images.This algorithm can quickly identify fire smoke information,send out an early warning signal in time,and warn staff to verify and process the early warning area.The main research contents and innovations of this article are as follows:1.Use the improved ViBe algorithm to extract suspected smoke.The inter-frame difference method,Gaussian mixed background modeling method,optical flow method,ViBe,etc.are commonly used algorithms for detecting video moving targets.Due to the strong realtime requirements of video smoke detection and high detection efficiency,and the ViBe algorithm is pixel level detection In line with the smoke detection in the real scene.In order to reduce the interference of non-smoke factors in video smoke,this paper proposes an improved ViBe algorithm,which restricts and judges the direction of smoke movement on the basis of the original ViBe algorithm,and minimizes the interference of non-smoke areas in video smoke.2.Detection of Suspected Smoke Regions Using Hybrid Deep Network Learning with Motion Context Information.The traditional smoke detection technology has certain influences on the real-time performance,false alarm rate,and missed detection rate of smoke detection due to human factors,but the deep learning technology avoids these factors.CNN has a powerful feature extraction function,which can extract the visual features and motion features of video images.At the same time,RNN can accumulate the motion context information of video images on the time stream,retaining the integrity of smoke features.Based on the characteristics of video smoke detection,this paper designs a hybrid deep network of motion context information based on indoor video smoke detection to detect smoke in the spatial domain,motion domain,and time domain.Finally,this paper gives a comparative analysis of the experimental results of the algorithm,and uses four commonly used performance indicators: the algorithm's response time,correct rate,false positive rate,and missed detection rate to measure the performance of the entire algorithm.The experimental results show that the proposed The algorithm has the advantages of good real-time performance and high detection accuracy.