Forest Fire Prevention Monitoring Context Envirnment Platform Design And Implementation

In the face of extremely dangerous forest fires,the existing fire warning systems based on visible light and thermal infrared are sensitive to temperature,and alarms are immediately triggered when fire occurs.However,the forest scenes are more complicated.Detecting fire based with the thermal infrared principle is likely to arouse false alarms.In order to drop the false alanm rate,this thesis designs a method of combining panoramic video stitching with semantic annotation and introduces the concept of contextual environment.This thesis constructs a platform for forest fire monitoring context environment.The work done in this thesis is as follows:(1)In order to solve the problem of the uncertainty of local image information,this thesis uses the output forest scene panorama to provide the preconditions for the subsequent construction of the context environment.The forest panorama required is a mosaic of image sequences captured by the camera.The specific process is to generate a panorama in accordance with the process of image preprocessing,detection and extraction of feature points,and matching and fusion of adjacent images after forest video images are acquired.Considering the quality and speed of image stitching,this thesis improves the stitching algorithm based on SIFT feature points,and adds parallel optimization processing in the three processes of image preprocessing,feature point detection,and image matching.(2)In view of the complexity of forest scenes and the high rate of false alarms in fires,this thesis uses image semantic annotation in panoramas to build a context model.And then uses multi-source verification for suspected warning messages issued by suspected fireworks to remove as much as possible.The specific process is to acquire the semantic databases of different scenes after the preprocessing of the panorama using the methods of manual annotation and SVM cluster annotation.Next current warning monitoring frame match the panorama.If the warning information passes through the scene in the semantic database,which include name,scope,and behavior rules,the system can eliminate the current alarm.Otherwise,alarm information will pass.This multi-source verification method improves the accuracy of forest fire warnings.In this thesis,the context of forest fire prevention monitoring is applied to specific forest scenes through the above two main processes.This system eliminates many alarm information in suspected fire areas,reduces false alarm rates for forest fires,and increases forest fires.Early warning monitoring accuracy.