FOV PREDICTION FOR 360 VIDEO STREAMING IN VIRTUAL REALITY

Abstract

In the transmission and review of 360 degree videos on VR devices (HMDs) initiate numerous specialized difficulties. 360 degree video are omnidirectional spherical video and on an approximate 4~6 times high resolution then normal 2D videos of same quality. Since 360-degree video are round video so having omnidirectional perspective of the scene. However human having Field of View (FOV) approximately 90 ~114 degree so only 1/4th of the part is viewed at time. Moreover, VR devices require to respond within 11ms to head movements of user to show the next Field of View (FOV).

In this task I research the issue of foreseeing the Field-of-Views (FOV) of client's viewing 360° video utilizing VR gadgets ahead of time. Existing and mostly present solutions while watching the 360 video first find out the current orientation and FOV of user’s and then request for high quality of video pertaining to FOV region and rest for low quality which induce bad user experience for some time called as latency to adapt the high quality.

I am developing solution to predict FOV in advance that simultaneouslyinclude content related historical data and sensor data to predict the viewer’s Field of View( FOV) in advance. The sensor-related features include VR devices orientation sensor like magnetometer and accelerometer, while the historical data include the complete data of previous watching orientation of video by different users. Based on historical data and sensor data, I shall train the system and validate design alternatives. Which will help in identify the better design to reduce the view latency and bandwidth required for 360 Video. Advantages of this solution will be (i) Lower bandwidth usage (ii) Low latency and (iii) short running time.