Panoramic Video Caching Strategy Based On Viewport Prediction Of Motion Content

Panoramic video,As a new type of immersive media not only brings immersive experience to the audience,but also puts forward the transmission demand of high bandwidth and low delay to the network.At the same time,with the promotion of panoramic VOD applications,there are more and more highly dynamic content such as sports and performing arts.As the interest points of sports video users are more dispersed and difficult to predict,the traditional edge cache framework is difficult to efficiently store the key interest blocks of users.Therefore,it is an urgent problem to study the view area prediction and related cache of highly dynamic panoramic video.For the application scenario of motion panoramic video on demand,a motion detection-based view area prediction and edge caching framework for panoramic video is proposed,which includes motion object detection and tracking view area prediction and edge caching.In order to obtain the motion vector information of panoramic video,this thesis proposes a visual area prediction algorithm based on motion vector field,which solves the problem of the same object moving across the region and the object separated by the projection frame edge in panoramic video,and realizes the extraction and tracking of moving objects in panoramic video.In order to predict the user’s visual field,this thesis proposes a visual field prediction algorithm based on motion vector field,by introducing motion object in the visual field prediction network vector field,achieved to contain user depending on the effective forecast of objects.Finally,for sports panoramic video,this thesis puts forward a kind of edge caching algorithm based on motion object area forecast.In this thesis,the MMSys 17 panoramic video dataset of Tsinghua University and MMSys 18 panoramic video dataset of Nantes University are used for experiments.The results show that :(1)compared with the existing PARIMA method,the number of detected objects in the proposed motion detection method increases by 1-2 frames per frame on average,and the tracking accuracy of the same moving object increases by 10%;(2)Compared with the LSTM method without motion detection,the accuracy of the region prediction method combined with motion detection improved by about 8%;(3)Compared with the traditional caching algorithm,the edge caching algorithm based on the prediction of the viewing area of the moving object improves the cache hit ratio by 6%,the maximum brightness peak signal to noise ratio of the video is improved by 1d B,and the consumption of return bandwidth is reduced by 28%...