Monitoring video target real-time query method based on edge cloud convolutional neural network cascading

Abstract

The invention discloses a monitoring video target real-time query method based on edge cloud convolutional neural network cascading, and belongs to the technical field of computer vision pedestrian re-identification. An edge end server extracts a moving target from the monitoring video, the lightweight CNN preferentially identifies the target, the target which cannot be judged is uploaded to the cloud server, and secondary identification is carried out by using the high-precision CNN. And meanwhile, an edge cloud task scheduling algorithm and a self-adaptive adjustment mechanism of threshold parameters are set, so that the performance of a monitoring video target query task can be effectively improved. Compared with a monitoring video target query method only using a cloud server, the invention has the advantages that the bandwidth consumption from an edge end server to the cloud server is remarkably reduced, the cost is reduced, the average query time delay and the fluctuation of thequery time delay are remarkably reduced, the real-time requirement is met, and meanwhile, the relatively good query accuracy is kept; compared with a monitoring video target query method only using anedge end server, the query accuracy is significantly improved.

Description

technical field [0001] The invention belongs to the technical field of edge computing / deep learning, and in particular relates to a real-time query method for surveillance video targets based on edge-cloud convolutional neural network cascading. Background technique [0002] With the widespread deployment of surveillance cameras, surveillance video data is growing explosively, and surveillance video analysis methods that rely on manual screening can no longer meet actual needs. On the one hand, the speed is slow, and it is difficult for manual screening methods to process a large amount of video data in real time; on the other hand, the cost is high, and massive video analysis brings huge labor costs. Therefore, with the development of deep learning, especially the development of convolutional neural network (CNN), automatic processing and analysis of surveillance video data is becoming more and more popular with the help of computer resources and video analysis algorithms. ...

AI Technical Summary

Problems solved by technology

However, uploading a large amount of surveillance video to the cloud server brings huge network bandwidth requirements and data transmission delays. It is difficult to meet the new real-time requirements of the surveillance video target query task, and it also brings a huge burden to the cloud server.

At the same time, most edge devices have limited resources and can only deploy lightweight CNN models (such as MobileNet), but cannot guarantee reliable query accuracy

Images