Video Compression Task Offloading For Unmanned Systems Based On Edge Computin

The ultra-low latency transmission,high speed broadband and ubiquitous access points greatly alleviate the last mile problem of streaming media live broadcasting services.However,before transmission,the complex compression coding of high-resolution video is still inevitable,which leads to the overload of light video collectors.As a promising solution for computing intensive and latency sensitive applications,edge computing has outstanding advantages in computing offload.It aims to enhance the distributed resources at the edge of the network and quickly deploy short-range services for mobile users.In the previous work,there have been a large number of successful use cases for task offloading based on edge computing,including augmented reality,face recognition,Internet of things analysis and so on.Aiming at the video encoding and transmission scenarios of UAVs with limited weight and power and small short-term changes in motion trajectories,this paper proposes a UAV video encoding framework combined with edge computing to realize the transfer of UAV video encoding loads to edge nodes.The specific works are as follows:(1)On the edge node,deep learning is used to detect objects in the obtained historical frame images,and segment the background areas in the images that move with the movement of the UAV camera and the foreground areas that may have autonomous motions,such as people,cars,etc.The segmentation information is transmitted to the UAV for encoding future frames of the video,and the segmentation information is transmitted to the client along with the video,which indirectly realizes the offloading of the client’s target detection task.(2)On the UAV,use the frame closest to the current frame to be coded that can obtain the foreground and background segmentation information from the edge node as the reference frame,and divide the current coded frame with the help of the segmentation information of the reference frame.For the foreground area of the current coded frame,bidirectional search is used to obtain the motion vectors,and for the background area,global motion estimation with lower computational complexity is used..The experimental results show that the proposed joint video coding scheme of UAV and edge nodes can greatly reduce the coding complexity of video encoder on UAV with little bit rate loss,and obtain better target detection results on the client.