Research On Fire Detection Technology Based On Video In Complex Scene

Fire is one of the most common and harmful disasters in human production and life,timely and effective fire warning technology has become the focus of research in recent years.With the popularization of monitoring equipment,the development of computer vision technology and the need of people’s production and life,the method of realizing fire alarm by analyzing video captured by monitoring equipment has gradually become one of the innovative directions of fire detection technology.Compared with traditional fire detection technology based on sensor,fire detection technology based on video content analysis does not need to install additional fire sensors.Fire detection can be realized by analyzing surveillance video,it can avoid some false alarms and missed alarms caused by sensor sensitivity;at the same time,it has the advantages of simple installation,wide detection range,not easy to be affected by the environment,convenient fire archiving,strong expansibility,high accuracy and real-time,it is suitable for mountainous forests,tunnels,warehouses and other space is relatively open and not suitable for the environments that are not suitable for installing traditional fire sensors.Based on the above advantages,video fire detection technology has become a hot spot of current research.With the development of artificial intelligence,computer vision and video processing technology,more and more efficient fire detection technology has been proposed.Video-based fire detection technology is mainly based on the analysis of video content to extract flame features,target recognition to achieve fire detection.However,due to the complexity and randomness of the natural environment,the irregular shape and movement of the flame,the current fire detection methods based on computer vision technology still has some disadvantages,such as high false alarm rate,slow detection speed and weak anti-interference ability.Aiming at the existing problems of current video fire detection technology,this paper proposes a video fire detection method that bases on target extraction,multi-target tracking and machine learning.Firstly,moving targets in video are extracted by moving detection algorithm of Gaussian mixture model.Secondly,fire targets are screened by color model combined with HSI and RGB,and suspected flame targets are obtained.Finally,flame and smoke are recognized by deep learning method to realize fire detection.Aiming at the motion characteristics of fire targets,a multi-target tracking algorithm based on inter-frame target distance matching is used to realize the stable tracking of suspected fire targets.In order to adapt to the task of fire detection in complex scenes,convolution neural network is used in the fire recognition module.Based on AlexNet network structure,the network model with the ability of flame and smoke recognition is obtained by fine-tuning training.Finally,the experiment proves show that the video fire detection technology proposed in this paper can effectively identify the fire areas in various complex scenarios,and eliminate the interference of suspected fire objects,meet the needs of real-time fire detection,and has good expansion and transplantation capabilities.