Efficient Coding And Transmission Scheme For Panoramic Video Based On User Viewing Behavior Prediction

Panoramic video can provide 360-degree video content for users,bringing an immersive viewing and interactive experience.However,the network bandwidth consumed by transmitting a high-quality,high-resolution panoramic video from the cloud server to the user is extremely huge,and because only a part of the view area can be viewed by the user at a certain moment,the cloud server sends a full-view high-quality panoramic video is essentially a waste of bandwidth.Therefore,this thesis has conducted in-depth research on the efficient encoding and transmission scheme of panoramic video.On the one hand,this thesis uses the feature of independent coding and decoding of Tile in the HEVC coding standard to propose a multi-quality Tile selection scheme,predicts the user's gaze point movement route based on a linear regression model,and combines quality deviation feedback to select different encoding quality for tiles in different regions of the panoramic video and finally transmits them.Compared with the full-view transmission single-quality Tile scheme,the scheme in this thesis can save an average of 12.16% coding rate under the same objective quality,and the highest sequence can save nearly 30%.On the other hand,the multi-quality Tile encoding and transmission scheme needs to obtain the real user's viewing behavior in real time and make predictions,so the coding performance is closely related to the user's viewing behavior.This thesis further explores how to predict the degree of interest in different areas of the panoramic video of virtual users without obtaining the gaze point of the real user.The virtual user viewing gaze point probability distribution prediction model MDNP based on the mixed density network proposed in this thesis has better prediction performance than the model DHP,which predicts the gaze point of virtual users based on the saliency of video content.The probability distribution map predicted by the scheme in this thesis improves the significance of the actual viewing path of batch users by 0.21.According to the predicted probability distribution,this thesis further establishes the mapping relationship between the viewing probability value and the quantization parameter to implement the non-uniform quality distribution coding of panoramic video.Encoding based on the prediction results of MDNP model and DHP model,the scheme in this thesis can save an average code rate of 0.72% under the same objective quality compared with the scheme based on DHP model.In addition,the two efficient coding schemes implemented in this thesis are compared.Compared with the multi-quality Tile coding and transmission scheme,the non-uniform quality distribution coding scheme based on the probability distribution of viewing fixation can save an average of 33.67% under the same objective quality.Finally,the full text is summarized and prospected.The advantages and disadvantages and application scenarios of the multi-quality tile coding and transmission scheme and the nonuniform quality distribution coding scheme based on the probability distribution of viewing gaze points are discussed.