Research On Fire Detection And Early Warning Based On Computer Vision

Fire has the characteristics of suddenness,great damage,and difficult to extinguish.Therefore,most modern buildings are equipped with sensors such as temperature sensors and smoke sensors to detect and deal with abnormal conditions as soon as possible.However,with the rapid development of the national economy,fire is used more and more frequently in industry and life,modern buildings and electrical equipments become more complex,these changes make the existing sensor detection equipment increasingly difficult to cope with the new fire situation.Therefore,the fire protection problem is becoming increasingly severe.Under this background,based on the computer vision technology,the problem of fire image information extraction,fire detection and early warning is studied and improved in this thesis,and based on this research,a fire-fighting system is built to minimize the threat of fire.The main work of this thesis is as follows:(1)The OpenCV toolkit is used to extract the fire information in the image before the fire detection.Through the suspected fire area extraction and fire feature extraction,a data set can be used for classification is gotten.Aiming at the problem that the existing RGB flame color model is easy to be interfered by brightness,the HSV and YCbCr models are used to separate the brightness and color information,and the rules of the fire color model is improved.Improved flame color model combines denoising preprocessing,GMM moving area extraction and morphological processing to make extraction of the suspected fire area more accurate;by analyzing the visual characteristics of the fire and its influence on the detection results,an appropriate amount of fire characteristics with good anti-interference effect and small coupling are selected for extraction method designing.The extracted fire features provide a basis for the fire detection later.(2)Aiming at the problem of the current video fire detection research that attaches importance to flame detection but ignores smoke and smoldering fire detection,an improved flame and smoke concurrent detection method is designed based on SVM tuning and threshold method.Through the SVM kernel function selection,model parameters optimization and training for SVM,it can be used for flame detection;through the improvement of the traditional threshold discrimination method,it can solve the shortcoming of weak generalization ability and the improved method is used for smoke detection;through the analysis of the existing problems in the fire early warning mechanism,an improved fire early warning mechanism is proposed.The experiment results show that the proposed fire detection algorithm can effectively improve the detection accuracy,reduce the false alarm rate and reduce the response time.(3)A fire-fighting experiment system based on computer vision and intelligent trolleys is built.The system consists of two subsystems,the fire detection and early warning system and the fire cooperative rescue platform,which together implement the fire detection,early warning and rescue functions.The fire detection and early warning system is developed by C ++ MFC program integrated with OpenCV,which can perform real-time fire detection on the video surveillance site and local video.After a fire occurs,it will perform detection background warning and remote SMS warning at the same time;based on the smart car and Andriod App,the fire cooperative rescue platform can be called by the fire commander and the trapped personnel.The smart car carries the required modules to enter the fire scene to complete the rescue tasks such as reconnaissance,positioning,and delivery of goods,so as to ensure the safety of firefighters and improve the efficiency of rescue.