Geo-referenced Video Coding Optimization Method Based On Attention

With the development of video GIS and Internet technology, the applicationbackground and function of Geo-Video have been immensely expanded, whichaccompany with the constantly improvement of video transmission service andcontinuous complexity of pathway requirement. Although recently the universal videocoding methods based on human visual feature selection optimization could relieve thiskind of pressure to some extent, Geo-Video has quite a lot of unique characters of thegeographic information science area. There still remain many problems in applying thecompression strategies of the existing coding standard directly to the Geo-Video.Therefore this thesis brought the Human visual attention mechanism into thegeographical spatial cognition, and proposed the video GIS visual attention calculationmodel. By adopting that to calculate the visual attention objects and visual attentionintensity, and get the frame attention and intra-frame visual redundancy objects of thevideos. Then achieve the compress optimization of geography video encoding, byemploying the frame attention and intra-frame visual redundancy objects to guide therate control strategy during the coding process.Firstly, this thesis analyzed the characteristics of the current video codingtechnology combined with video coding framework, and discussed the three keytechnologies Geo-Video encoding optimization involved, that is the human visualperception, the geospatial cognition and the geographic video coding. Secondly,combining with the characteristics of human visual attention mechanism and video GIS,this thesis provided the definition of video GIS visual attention aimed at the Geo-Videoencoding optimization. And on this basis, it analyzed the video image motion featureextraction technique, and deeply discussed the description method of the characteristicsof the Geo-Video image motion. Thirdly, it proposed the video spatial patterndescription method that combined with the geospatial cognitive theory and the humanvisual perception theory, which can be used to reflect the spatial distribution and theconfiguration of the visual elements in the Geo-Video. Fourthly, this paper combiningwith the video motion characterization method and the spatial pattern descriptionmethod to establish the model that can effectively calculate the Geo-Video frame visualattention content area and inter-frame visual attention intensity, and use the model tooptimize the rate control strategy during the video coding stage, thus achieving the purpose of improving the efficiency of video compression. Finally, this thesis took theon-board street view video re-encoding system as an example to verify the optimizationtheory this thesis proposed, and initially realized encoding compression optimization ofthe on-board street view video.Through a comprehensive analysis of the results, it shows that the proposedmethod can effectively wipe off the time and space redundancy of Geo-Video, optimizethe video compression results and provide the novel thought for Geo-Video encodingefficient transmission.