摘要
提出了一种高效的人体视频的无线传输方法,称为人体动作视频语义传输(AAST)系统。在当今的应用场景中,H.265视频编码和5GLDPC信道编码的组合已成为视频传输的常见方法。然而,在传输效率方面,尤其是在特殊情况下,仍存在相当大的提升潜力。受到最近人体动作视频研究的进展启发,提出了AAST系统。AAST系统能够以高效的方式传输关键区域的语义信息,并在接收端重建原始视频。鉴于关键区域运动语义信息内在的相关性,AAST系统采用非线性分析变换将运动信息映射到潜在空间。然后通过信源信道联合编码,将运动信息的潜在表示传输到接收端。在该系统中,利用潜在表示的先验信息来估计关键区域语义信息的重要性。这有助于在信源信道联合编码中实现自适应码率控制,从而带来显著的编码增益。大量的实验证明了AAST系统相对于其他有竞争性的视频传输系统具有更卓越的性能。值得注意的是,与H.265视频编码和5GLDPC信道编码传输方法相比,在达到相同的感知指标时,AAST系统最高可以节省约70%的信道带宽,展现了其在视频无线传输中的鲁棒性。
An efficient wireless transmission method for human body motion videos is proposed,termed articulated animation semantic transmission(AAST)system.In contemporary scenarios,the combination of H.265 video coding and 5G LDPC channel coding becomes a common approach for video transmission,However,there is still considerable potential for improvement in terms of transmission efficiency,especially in special scenarios.Inspired by recent advancements in body motion video studies,the AAST system is introduced.The AAST system can efficiently transmit semantic information of key regions and reconstruct the original video at the receiver.Given the inherent correlation of motion semantic information in key regions,the AAST system employs a nonlinear analysis transformation to map the motion information into a latent space.This latent representation is subsequently transmitted to the receiver via joint source-channel coding.This system leverages prior information of the latent representation to estimate the importance associated with the semantic information in key regions.This facilitates adaptive rate control in joint source-channel coding,leading to notable coding gains.Comprehensive experiments validate the superior performance of the AAST system compared to other competing video transmission systems.Remarkably,compared to the transmission method involving H.265 video coding and 5G LDPC channel coding,the AAST system can save up to approximately 70%of channel bandwidth when achieving the same perceptual metrics,underscoring its robustness in wireless transmission of videos.
作者
王沛辰
岳伟杰
党天健
戴金晟
牛凯
WANG Peichen;YUE Weijie;DANG Tianjian;DAI Jincheng;NIU Kai(The Key Laboratory of Universal Wireless Communications,Ministry ofEducation,Beijing University of Posts and Telecommunications,Beijing 100876,China)
出处
《移动通信》
2024年第2期63-69,共7页
Mobile Communications
基金
国家自然科学基金“语义通信基础理论与方法研究”(62293481)
国家自然科学基金“语义驱动的工业互联网原生智简组织理论”(92067202)。
关键词
语义通信
人体动作视频
信源信道联合编码
变分模型
semantic communications
human body motion video
joint source-channel coding
variational modeling
