Eye-tracking Based Video Streaming System with Fast Quality Switching

Abstract: While more people than ever consume video content using mobile devices, high resolution video requires high bandwidth, which is a problem in current wireless networks. The requirements can be even higher for spherical video for virtual reality devices, because only a portion of each frame is shown on the screen. Thus, in order to deliver acceptable resolution for the viewport, frames must be sent in a much higher resolution.This thesis proposes a streaming system that takes advantage of characteristics of the human visual system to reduce the bandwidth necessary to deliver video over low latency networks. The system works by splitting videos spatially into tiles and using eye-tracking to select which sections should be sent in higher quality. A method to quickly switch the quality in the tiles is proposed and validated. A testbed that allows investigating the system parameters and its performance was built, and experiments show that with latencies possible for 5G networks, bandwidth consumption can be reduced by as much as 75% for conventional video and 83% for spherical video while maintaining an acceptable quality of experience