Geographic Video Target Information Extraction And Visualization Method

Surveillance video contains rich spatiotemporal information,but the information is mainly expressed in the image space,which is independent,difficult to coordinate,and difficult to analyze in space.GIS manages spatiotemporal data through a strict geospatial reference system,and builds a framework that is easy to integrate multi-source geospatial data,with the advantages of at-a-glance,locatable,measurable.Video GIS,which integrates video surveillance system and GIS,can give full play to the advantages of both,effectively organize scattered and unorganized video information,and achieve efficient control of the overall situation of the surveillance area.In practical applications,the viewing angle and scene information obtained by a single camera have an upper limit,which often cannot cover the field of view of the entire area.In order to achieve a comprehensive monitoring effect,multiple fixed cameras are usually deployed in the monitoring area or PTZ cameras are used for repeated scanning.However,in scene monitoring,it is still necessary to view each camera one by one,and the monitoring images are numerous and isolated from each other,making it difficult to establish the overall situation of the monitoring scene.Video GIS can integrate the spatiotemporal information of surveillance video into the map,reducing the difficulty of comprehensive analysis,but the existing research mainly focuses on the processing of single camera video.The fusion of multi-camera image/video mostly stays in the image space,and less attention is paid to the fusion of multi-camera spatiotemporal information in geographic space.Based on the above research background,the main research contents and results are as follows:(1)A multi-camera spatiotemporal information fusion method considering imaging pose is proposed.Aiming at the video spatialization of fixed cameras,based on the estimation of homography matrix,an image control point selection method is proposed,which improves the accuracy of spatial positioning.Based on the deep learning method,two target detection models of people's front-view and rear-view are constructed to improve the accuracy of target detection.Using the target selection method of overlapping viewports of multiple cameras in geographic space,the selection of the only target under the overlapping viewports is realized.(2)A method for spatialization of PTZ camera video based on cylindrical panorama is proposed.Aiming at the spatialization of PTZ camera video with flexible posture,a cylindrical panorama and geospatial mapping model is proposed,which solves the problems of spatialization of PTZ camera and difficulty in fusion of targets after geographic mapping with the help of cylindrical panorama.After segmenting the spliced image,target detection is performed,which improves the accuracy and speed of target detection while ensuring the integrity of the target.(3)A method for visualizing geographic video target information is proposed.Based on the extraction of spatiotemporal information from surveillance video,a visualization method of geographic video target information is proposed.The extracted geographic video targets are abstracted into map symbols,graphic symbols are designed according to the target state,and the time attribute is taken as a part of the symbol structure.Using the heat map visualization method,the spatial and temporal distribution of the target in the spatial area can be displayed intuitively.