TY - GEN
T1 - Can Accurate Future Bandwidth Prediction Improve Volumetric Video Streaming Experience?
AU - Khan, Muhammad Jalal
AU - Bentaleb, Abdelhak
AU - Harous, Saad
N1 - Publisher Copyright:
© 2021 IEEE
PY - 2021
Y1 - 2021
N2 - Recently, the advancements in technologies have enabled volumetric media techniques to capture, encode, decode, and render videos in six degree-of-freedom (6DoF) in order to make the objects highly immersive, interactive, and expressive within the scene. This is enabled by using multiple cameras around the object(s). However, streaming 6DoF videos require a huge bandwidth and computational processing. As the end-user focuses on viewport-scenes, a large portion of the consumed bandwidth is mainly introduced due to unseen video-scenes. To fill this gap, it is imperative to predict the future head-movement (future viewport) of end-user in order to avoid the waste of network bandwidth and reduce computational processing power. In this paper, we propose a holistic architecture for the future viewport prediction using deep-neural-network (DNN)-based model. Specifically, our solution uses residual long-short-term-memory (RLSTM) architecture for accurate future viewport prediction. We confirm the effectiveness of our solution through trace-driven streaming experiments using a popular public dataset over four categories of DNN models: linear, dense, convolutional, and long-short-term-memory (LSTM). Experimental results show that our solution is able to achieve the lowest possible mean absolute error of ∼ 0.01 compared to its competitor.
AB - Recently, the advancements in technologies have enabled volumetric media techniques to capture, encode, decode, and render videos in six degree-of-freedom (6DoF) in order to make the objects highly immersive, interactive, and expressive within the scene. This is enabled by using multiple cameras around the object(s). However, streaming 6DoF videos require a huge bandwidth and computational processing. As the end-user focuses on viewport-scenes, a large portion of the consumed bandwidth is mainly introduced due to unseen video-scenes. To fill this gap, it is imperative to predict the future head-movement (future viewport) of end-user in order to avoid the waste of network bandwidth and reduce computational processing power. In this paper, we propose a holistic architecture for the future viewport prediction using deep-neural-network (DNN)-based model. Specifically, our solution uses residual long-short-term-memory (RLSTM) architecture for accurate future viewport prediction. We confirm the effectiveness of our solution through trace-driven streaming experiments using a popular public dataset over four categories of DNN models: linear, dense, convolutional, and long-short-term-memory (LSTM). Experimental results show that our solution is able to achieve the lowest possible mean absolute error of ∼ 0.01 compared to its competitor.
KW - 6DoF
KW - Adaptive video streaming
KW - DASH
KW - HAS
KW - HLS
KW - Viewport prediction
KW - Volumetric videos
UR - http://www.scopus.com/inward/record.url?scp=85125642710&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85125642710&partnerID=8YFLogxK
U2 - 10.1109/IWCMC51323.2021.9498691
DO - 10.1109/IWCMC51323.2021.9498691
M3 - Conference contribution
AN - SCOPUS:85125642710
T3 - 2021 International Wireless Communications and Mobile Computing, IWCMC 2021
SP - 1041
EP - 1047
BT - 2021 International Wireless Communications and Mobile Computing, IWCMC 2021
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 17th IEEE International Wireless Communications and Mobile Computing, IWCMC 2021
Y2 - 28 June 2021 through 2 July 2021
ER -