Gaussian-scale-space-based Fire Images Local Features Extraction And Proactive Fire Identification Recearch

With the rapid economic development and improvement of living standards,the demand of security for living,property and life safety is increasingly urgent.In recent years,the government made ignificant investments in the field of public safety,such as The Construction of Wisdom and Peace City,Construction of the Internet.And with implementation of the famous projects,Beijing Olympic Games,Shanghai World Expo and Guangzhou Asian Games,Shenzhen Universiade and so on,security system in Public domain promoted with each passing day.The field of security monitoring ushered the times of High-Definition,informatization and intelligence.Video fire detection is a kind of technology for fire recognition intelligently and early warning using video security-monitoring system.It is the part of realizing intelligence most directly for the security systems,and also the most important measures for life and property safety against fire.However,the current technology of video fire detection can not meet the rapid development of electronic hardwares.Characteristics studies focused on some of the integral and global underlying features of the fire image,no in-depth study of the local structure and natural law.Research in the field among scholars is relatively fragmented,lack of uniform standard database and information sharing platform.It is not conducive to promote the sharing of resources and academic exchanges.The Modem 1080P HD monitoring system provides high-resolution images,so the picture is clear and fluent,rich in local details,Large amounts of data supply new data mining space to develop new feature models,new identification methods and functions.To solve these problems,this paper developed a simulated fire HD video image data acquisition platform,and established a standard database shared within field.Study the local feature of high-definition images of fire,designed the program to real-time process high-definition streams,and developed new fire identification methods base on high-definition video.specifically,this paper studied these contents as the following aspects:(1)Designed fire experiment platform for HD image and video acquisition and experimented to establish standard fire databases of HD image and video.According to machine learning theory,we designed fire experimental data acquisition devices,including collection experiments for training data and experiments for testing data.And designed working conditions of the experiments,including the use of visible light,infrared,thermal infrared image acquisition device,open space,confined space,natural conditions,forced ventilation,fire and smoke experiments under different light conditions.And developed database standards,built standard fires database of High-definition image and video for training and testing.(2)Studied the local features of high-definition images of fire smoke and flame,established the local feature models of flames and smoke,and developed high-definition-video-based fire detection algorithm.This paper used local binary pattern(LBP),Gabor operator,corner detection operator,SIFT operator and other local feature descriptors to extract local texture features,the frequency-domain characteristics,corner feature,and the key point invariant features of flames and smoke,and the featrue-based models were trained using the standard database.For high-stream videos of H.264,selective pretreatment method of weighting image block,instead of the traditional approach based on the whole image,are targeted to process the target area,greatly reducing the amount of data processing and improving the data processing efficiency.Adaboost classifier was used to learn and train,and developed high-definition video fire detection algorithm based on local features.The testing results indicated that the recognition accuracy rate,sensitivity level and respond speed were improved.Solved the limitations of traditional features,long time and close recognition,small range of applications and other problems of traditional algorithms.(3)Studied the scale invariant featrues of the local characteristics of fire images,proposed proactive HD-video-based fire identification method using digital foucus.Gaussian scale-space theory was brought in to build a fire Gaussian scale-space image expression,and trained the scale invariant local feature model.For high-definition images,sacrifice some resolution to stretch the fire suspicious target area,zoom in focus.Used the scale invariant features to recognize fire ahead and actively.Developed a proactive fire identification method with a faster response and a higher level sensitivity.