车联网中视频语义驱动的资源分配算法  

Video semantics-driven resource allocation algorithm in Internet of vehicles

作　　者：陈九九 冯春燕[1] 郭彩丽[1] 杨洋[1] 孙启政 朱美逸 CHEN Jiujiu;FENG Chunyan;GUO Caili;YANG Yang;SUN Qizheng;ZHU Meiyi(Beijing Laboratory of Advanced Information Networks,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学北京先进信息网络实验室,北京100876

出　　处：《通信学报》2021年第7期1-11,共11页Journal on Communications

基　　金：中央高校基本科研业务费专项资金资助项目(No.2021XD-A01-1);国家自然科学基金重大研究计划重点资助项目(No.92067202);北京市自然科学基金资助项目(No.4202049);北京邮电大学(济南)工业互联网研究院项目(No.201915001)。

摘　　要：针对车联网中视频语义理解等智能计算业务需求下传统资源分配方式不再适用的问题,研究了视频语义驱动的资源分配算法。首先,以目标检测任务为例,提出视频语义驱动的资源分配指导模型并给出模型参数的求解算法;其次,构建了车联网场景中视频语义驱动的资源分配优化问题,将该问题转化成凸问题并利用凸优化算法求解;进一步,为降低凸优化算法的复杂度,提出了基于强化Q学习的资源分配算法;最后,仿真验证了所提资源分配算法的性能优势。Aiming at the problem that traditional resource allocation methods will no longer be applicable,with the de-mand of intelligent computing services such as video semantic understanding in Internet of vehicles,the video semantic driven resource allocation algorithm was studied.First of all,taking the object detection task as an example,a semantic driven resource allocation guidance model for video was proposed and an algorithm for solving model parameters was given.Secondly,an optimization problem of resource allocation driven by video semantics in Internet of vehicles was constructed,which was transformed into a convex problem and solved by convex optimization algorithm.Furthermore,in order to reduce the complexity of the convex optimization algorithm,a resource allocation algorithm based on reinforce-ment Q learning was proposed.Finally,the performance advantages of the proposed algorithm are verified by simula-tions.

关 键 词：资源分配 车联网 视频语义 目标检测 强化学习 

分 类 号：TN929.5[电子电信—通信与信息系统]