Content-Aware Optimization of Tiled 360° Video Streaming Over Cellular Network

The proliferation of 360° video content and easy availability of head-mounted devices propelled the popularity of 360° video streaming on various platforms, including smart-phones. However, the high bandwidth requirement of 360° video streaming impedes its further growth, specifically in bandwidth constraint cellular networks. Some viewport adaptive 360° video streaming methods have been proposed recently to reduce the bandwidth requirement by spatially reducing the transmitted data. However, these methods require accurate prediction of the future viewport, and a prediction error can significantly impair the user’s Quality of Experience (QoE). In this work, we took a different approach and propose a content-aware solution for tiled 360° video streaming that does not require viewport predictions. In this work, we exploit the fact that the background scene does not contain prominent details and can be transmitted at a lower bit rate without affecting the user’s QoE. The proposed approach uses the saliency map to identify prevalent tiles of the 360° video through machine learning-based classification using the Kernel Density Estimation (KDE). First, the tiles are classified based on their saliency information, and then bitrates are assigned to the tile clusters. The results show that the proposed approach reduces the bandwidth requirement by 50% while delivering the equivalent user’s QoE.