Track: a Multi-Modal Deep Architecture for Head Motion Prediction in 360<sup>°</sup> Videos

Rondón, Miguel Fabián Romero; Sassatelli, Lucile; Pardo, Ramón Aparicio; Precioso, Fŕed́eric

doi:10.1109/icip40778.2020.9191331

Cited by 8 publications

(1 citation statement)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the latest FoV prediction models result in a low long-term prediction accuracy [67]. Rondon et al [68] reported that existing neural network models used for both content-based and contentindependent viewport prediction perform worse than a basic (last known) approach that simply uses the last viewing position for the next segment. Due to the extremely unpredictable viewing nature of the user, the basic idea is to stream more tiles than necessary to cover the actual viewing area.…”

Section: B Cfov Tiles Selectionmentioning

confidence: 99%

A Combined Field-of-View Prediction-Assisted Viewport Adaptive Delivery Scheme for 360° Videos

Yaqoob

Muntean

2021

IEEE Trans. on Broadcast.

View full text Add to dashboard Cite

Recently, 360 • or omnidirectional videos have become increasingly popular for both personal and enterprize use-cases. However, 360 • video streaming has very high bandwidth and processing requirements. State-of-the-art viewport-based streaming solutions lower these requirements by performing selective streaming based on long-term Field-of-View (FoV) prediction mechanisms. However, sometimes user movement is extremely unpredictable during some parts of the video, and applying these solutions adversely affects the overall quality of experience (QoE). This paper proposes a novel Combined Field-of-View tile-based adaptive streaming solution (CFOV) that improves end-user QoE for 360 • video streaming. CFOV performs interactive tile selection based on more accurate dynamical viewing area identification by combining the results of two FoV prediction mechanisms. It also employs an innovative priority-based bitrate adaptation approach that ensures improved bitrate budget distribution between different tiles. We evaluate the proposed solution with a comprehensive set of experiments involving four immersive videos, diverse tiling patterns (i.e., 4x3, 6x4, and 8x6), different segment lengths (i.e., 1s, 2s, and 3s), and 48 empirical head movement traces under different bandwidth settings. The evaluation employs a newly defined QoE metric specifically introduced to assess the streaming performance of 360 • videos objectively. The experimental findings show that, compared to alternative approaches, our proposed solution can achieve a higher viewport match and can significantly improve the user QoE for different watching behaviors and content characteristics.

show abstract

Section: B Cfov Tiles Selectionmentioning

confidence: 99%