Rich360-Live: A Practical Approach to Live Adaptive 360° Streaming via Content-Aware Optimization

초록

With the increasing demand for 360-degree live streaming, delivering high-fidelity immersive experiences under limited bandwidth remains a significant challenge. While viewport-adaptive streaming reduces bandwidth usage, it frequently suffers from motion-to-photon latency and visual artifacts due to viewport prediction errors. Conversely, existing content-aware approaches typically lack the computational efficiency required for real-time live scenarios. To address these limitations, we propose Rich360-Live, a practical end-to-end streaming system designed for real-time processing on standard consumer-grade hardware. Our approach introduces a Hybrid Importance Mechanism that synergizes automated visual tracking with directorial control, ensuring that semantically important regions are preserved. We employ a lightweight Axis-Aligned Mesh Optimization algorithm that drastically reduces computational complexity by converting the 2D grid optimization into a 1D parameterization. This enables effective non-uniform downsampling that maximizes Region of Interest (ROI) quality while preventing background degradation through a novel baseline quality constraint. Furthermore, to guarantee frame-accurate synchronization, we implement an in-frame parameter embedding technique that encodes grid parameters directly into video frames, effectively eliminating mismatches associated with separate metadata transmission. Extensive experiments and a user study demonstrate that Rich360-Live achieves superior visual stability and Quality of Experience (QoE), outperforming state-of-the-art methods such as MANSY and robustMPC by 20% and 39%, respectively, particularly in bandwidth-constrained environments.

키워드