多端交通视频分析任务卸载决策

Multi-client traffic video analysis task offloading decision

针对智慧交通中多点位监控视频分析时出现的计算任务繁重、设备之间资源利用不均衡的问题,提出一种基于云控制的视频分析卸载方案.首先,针对客户端算力不足而无法完成视频分析任务的问题,使用一种视频卸载框架,将部分视频分析任务切块卸载至云服务器处理;其次,针对服务器与多客户端之间的算力资源竞争问题,提出一种阶段优化卸载算法,平衡设备之间负荷,提高资源利用率;最后,针对不同点位的客户端需求不同的问题,在算法中加入精度和能耗偏好,满足不同客户端的需求.与其他卸载方案对比的实验表明,所提出方案能够更好地对视频分析任务进行合理分配,提高系统收益,并通过扩展实验验证所提出系统的扩展能力.

To address the problem of heavy computational tasks and uneven resource utilization among devices in multi-point monitoring video analysis in intelligent transportation,a cloud-controlled video analysis offloading scheme is proposed.Firstly,for the problem of insufficient client computing power to complete video analysis tasks,a video offloading framework is used to segment and offload some of the video analysis tasks to cloud servers.Then,a stage-optimized offloading algorithm is proposed to balance the load between servers and multiple clients,and improve resource utilization.Finally,to address the issue of different client requirements at different points,precision and energy consumption preferences are added to the algorithm to meet the needs of different clients.Experimental comparisons with other offloading schemes demonstrate that this scheme can better allocate video analysis tasks,improve system benefits,and the scalability of the system is demonstrated through extended experiments.