| Geographic video can show the real-time changes of geographic scenes and the interaction of dynamic objects,and has become a new way of geospatial expression.The advantages of high-definition,real-time and easy access of surveillance video and the features of accurate and intuitive geographic location expression,rich spatiotemporal information and outstanding visualization effect of GIS lay the foundation for the wide application of geographic video.Due to the complexity of camera perspective imaging technology and geographic scenes,it is easy to ignore the semantics in the video when conducting information extraction.The dynamic object in the video is mapped directly without extraction,which will result in serious deformation of the object.Existing research mostly pays attention to single camera for information extraction,and there are difficulties in fusion and analysis for multi-camera videos with overlapping fields of view.The fusion of automatic matching method mainly relies on image information,and ignores the role of geographic information in video fusion,so it is difficult to realize image fusion and object fusion from the perspective of geographic space.By means of collaborative analysis of video surveillance data and geographic information data,this paper extracts and analyzes video spatiotemporal data in geographical scenes,and explores the method of geographic video fusion and live broadcasting with semantic consideration.The research contents and main results of this paper are as follows:(1)A semantically sensitive orthographic video generation method is proposed.Based on GIS,the information of the camera video is extracted.The GMM and semantic segmentation model are used to extract the information of the playground in the ground area.By training the playground object detection model,the foreground object position and subgraph are extracted.The extracted video foreground and background information is converted from image coordinates to geographical coordinates through homography matrix,and orthographic video is generated.Compared with the traditional surveillance video,the method in this paper takes into account the geographic semantics and extracts information for the geographical entity of the playground,reduces the deformation of the object,and improves the visualization effect of the mapped video image and the object.(2)A multi-camera geographic video fusion method is proposed.In this paper,the algorithm of background fusion and foreground object fusion is proposed based on multi-camera videos and geographic information.For video background fusion,four indexes including distance,angle,area and length deformation ratio are considered to complete the fusion of two camera background videos.For the playground foreground object fusion,the three indexes of object confidence,distance and angle were considered to optimize the foreground object and realize the fusion of the foreground dynamic object.The results show that the multi-camera geographic video fusion method proposed in this paper can effectively realize the fusion of multi-camera geographic video,reduces the repeated video data of multiple cameras,and improves the availability of spatiotemporal information.(3)A geographic video live streaming prototype system was developed.Based on the above research results,a geographic video live streaming prototype system has been developed,which uses Web Socket technology and designs a C/S and B/S hybrid software framework suitable for Web GIS integration.The background information and foreground information of the extracted and fused video will be published and live broadcast,and real-time browsing will be conducted under WebGIS. |
PDF Full Text Request