Optimized Tile Quality Selection in Multi-User 360° Video Streaming

Abstract

360° videos and virtual reality (VR) are among the defining applications for a new era of immersive multimedia technologies. However, streaming high-quality 360° videos constitutes a challenge for service providers and network operators, especially in multi-user scenarios. The massive data rates and the unique features of omnidirectional videos necessitate the development of novel streaming techniques. In this paper, we propose an optimized multi-user tiled 360° video streaming framework. Specifically, we formulate the problem of tile quality selection in multi-user bandwidth-constrained communications and propose an algorithm that assigns the quality levels of transmitted tiles. The proposed framework is designed to maximize the perceived quality within the users' viewports by considering 360° video quality assessment (360° VQA) metrics and tile viewing percentages. To simulate our solution in practical settings, we employ a long short-term memory (LSTM) model to perform viewport prediction and make the assessment based on real-life viewing data. Simulation results of our proposed framework show significant improvement in the delivered viewports' quality and robustness against increasing the number of users.