Streaming video face detection acceleration method based on frame buffer queue

Abstract

The invention relates to a face detection acceleration method in streaming video processing based on a frame buffer queue. The method mainly comprises three parts, namely an edge real-time video processing system based on a frame buffer queue, a self-adaptive scaling factor setting method for different resolutions and a dynamic setting method for a human face target detection area. According to the invention, the frame accumulation caused by the network environment and the frame receiving time delay is reduced, and the streaming video processing performance is optimized. Besides, in order to improve the processing performance of the high-resolution streaming video, the network bandwidth resources and the power consumption of the edge server are considered for different streaming video resolutions, and the processing overhead is minimized while the video processing performance is ensured. In order to improve the face detection effect in the streaming video, a face detection area is dynamically set according to different video resolutions.

Description

technical field [0001] The invention relates to performance optimization of face detection in real-time streaming video processing, including server performance optimization and video processing performance optimization, especially for processing real-time video streams of video capture nodes in an edge computing environment and face detection acceleration methods. Background technique [0002] Cloud computing technology provides computing and storage to end users in the form of services, which reduces the cost of hardware infrastructure construction and operation and maintenance for users, and enables users to use various services in the cloud computing center to process massive video data. However, due to the explosive growth of video streaming data brought about by the Internet of Everything, video processing based on the cloud computing model has problems such as high network bandwidth requirements, poor real-time performance, and poor security and privacy protection. Th...

AI Technical Summary

Problems solved by technology

Because video is a complex data type, video transmission will consume a lot of network resources, and video processing will consume a lot of system resources, and the hardware resources of network edge devices are often limited. Therefore, edge servers need to meet their own performance. Maximize Video Processing Performance

[0004] In a data communication network, data packet loss may occur due to line failure, equipment abnormality, network congestion, etc.

In the streaming video surveillance network, in addition to the above reasons, frame accumulation may be caused by insufficient hardware performance of the stream processing system and software algorithm defects, which may cause decoding errors and cause frame loss.

Images