The Acceleration Method Of Video Processing For Edge Computing

With the deployment of a large number of HD network cameras,edge video analysis has played an important role in edge computing,but it still has many performance problems due to the limited computing power of many edge devices.For example,video processing of high resolution and frame rate are intend to lead the problem of frame accumulation easily.Therefore,a method is urged to reduce the accumulation of frames,so as to improve the real-time performance of the video analysis system.In addition,the diversity of video parameters also affect video processing.So,the video processing system needs to adjust the system parameters to ensure video processing performance adaptively for the video data with different parameters.Aiming at the above problems,this paper tries to solve as follows:First of all,a method is proposed to add a buffer(Frame Buffer Queue,FBQ)between reception and processing of frame to process buffered frames in parallel to solve frame reception delay and accelerate video processing.The experimental results show that the frame buffer queue can solve the frame dropping problem,while meeting the real-time processing of frames,reducing the system power consumption,and improving the edge computing ability to process real-time video data.Then,this paper analyzes the effect of image scaling factor on preprocessing of video with different resolutions,and proposes a dynamic setting scheme for the scaling factor.The scheme uses the system performance index(server power and memory usage)as the constraints on video processing performance index(face detection rate).Base on it,the scaling factor corresponding to the optimal face detection rate is considered as the target scaling factor corresponding to the resolution.The experimental results show that the scheme can reduce the system power consumption and memory usage for the videos with different resolutions,while guaranteeing the video processing performance.Finally,this paper explores the impact of target detection areas on the processing of video with different resolutions.Based on the fact that the ratio between the pixels in the target detection area and the pixels in the image is related to the shooting distance,but not the resolution,the region pixel point is obtained by the video resolution and the corresponding ratio and is taken as the axis center for the specific shooting distance.The region pixel point,which minus the axis distance and plus axis distance are the upper and lower limits of the detection area pixel respectively.With the help of optimal axis distance,which is obtained through experiments,face detection rate of the system and shortening the shorter processing time can be guaranteed.According to the mapping relationship among shooting distance,the ratio(the number of target area pixels/the number of image pixels)and axis distance,the system sets a reasonable target detection area automatically for images of different resolutions.As a result of this,sufficient face information can be obtained with the least video data.For video data with different resolutions,the above method can reduce the video processing time and improve the efficiency while ensuring the frame reception rate and face detection rate.In summary,the method proposed in this paper can improve the performance of real-time video processing in edge effectively,so it is promising to be applied to the performance improvement scheme of edge real-time video platforms.