Research On 360 Video Adaptive Transmission Algorithm Based On Viewpoint Feature

With the rapid expansion of the Virtual Reality(VR)market,relevant technologies have made great progress.360 video,as an important application of virtual reality,not only receives attention from the majority of users,but also poses challenges to network transmission.The common practice in the industry is to use tile-based transmission method.The server divides a 360 video into different tiles in space,and uses different bitrates to transmit tiles at different positions to reduce the redundancy of data during transmission.On the basis of tile-based transmission scheme,this paper proposes an adaptive transmission scheme of 360 video based on viewpoint feature.The scheme is divided into three stages:viewpoint position feature prediction,tile occurrence probability prediction and adaptive bitrate allocation.In the stage of viewpoint position feature prediction,this paper proposes a prediction algorithm of viewpoint position feature based on neural network autoregression model.It uses the user’s viewpoint information as the trend prediction basis and related viewpoint trajectory of other users as seasonal prediction basis to predict the future viewpoint coordinate.In the stage of tile probability prediction,this paper introduces the concept of viewpoint probability distribution feature to reduce the adverse impact of viewpoint position feature prediction error on user’s quality of experience(QoE).It uses Kalman Filter and viewpoint heat map to evaluate the viewpoint probability distribution,respectively.Then it uses maximum likelihood estimation method to combine the two distributions adaptively to ensure the robustness of the performance.Based on the viewpoint probability distribution and the shape of the field of view(FoV)in Equi-Rectangular Projection,the occurrence probability of tiles in FoV is obtained.In the stage of bitrate adaptive allocation,this paper classifies the throughput resource allocation problem as a knapsack problem.In the design of the algorithm,the aim of this paper is to reduce the time complexity of the algorithm and to solve the optimal solution of the problem.Finally,the algorithm is designed based on the idea of dynamic programming,so that it can get the optimal solution under the requirement of low time complexity.In this paper,the proposed algorithm are simulated,and other schemes are implemented respectively for comparison with ablation experiments.The final experimental results show that the viewpoint position feature prediction algorithm,the tile occurrence probability prediction algorithm,and the bitrate adaptive allocation algorithm proposed in this paper can all achieve better results,and improve the user’s QoE.