Edge-Cloud Video Analysis Frame Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video analysis methods in edge and cloud servers face challenges due to limited computational resources on the edge side, resulting in insufficient accuracy of cut-out images and subsequent video analysis on the cloud server.
Innovation Solution
A method that involves analyzing input image frames on the edge side using a lightweight model, estimating difference values between edge and cloud analysis results, and filtering to determine which frames to transmit to the cloud server for higher accuracy analysis, thereby improving overall video analysis accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If video is transmitted to cloud server via network, then video analysis can be performed with abundant computational resources, but video quality needs to be lowered due to bandwidth limitation
Solution Approach 1:
The system segments the video analysis task into two parts: edge-side analysis for real-time processing and cloud server analysis for high-accuracy analysis. Only selected frames are transmitted to the cloud server, while the edge device handles routine analysis locally, dividing the computational workload appropriately.
Solution Approach 2:
The edge device performs preliminary analysis on video frames before transmission to the cloud server. It predicts which frames would benefit most from cloud analysis by estimating the difference between edge and cloud analysis results, and only transmits those frames, avoiding unnecessary transmission of low-value frames.
2Productivity
If video analysis is performed in a distributed manner by edge and cloud, then computational resources are utilized efficiently, but it is difficult to determine which video frame needs to be transmitted to the cloud side
Solution Approach 1:
The system uses feedback from the edge device's analysis results to determine cloud transmission needs. The edge device estimates the difference between its analysis results and what the cloud server would produce, using this feedback to intelligently select which frames to transmit, making the distributed system self-regulating.
Solution Approach 2:
The edge device autonomously determines which frames to transmit to the cloud server without external control. It uses its own analysis capabilities to predict the value of each frame and makes self-directed transmission decisions, simplifying the system architecture.
3Ease of operation
If a monitoring terminal on edge side extracts cut-out images, then cloud server can receive processed images, but the terminal hard to properly extract cut-out image due to limited computational resources
Solution Approach 1:
The edge device acts as an intermediary that performs preliminary analysis and frame selection before transmission to the cloud server. It bridges the gap between limited local resources and the need for high-accuracy analysis by preparing and filtering frames appropriately.
Solution Approach 2:
Instead of attempting to perform all analysis operations locally (excessive action), the edge device performs only the necessary preliminary analysis and selection operations (partial action), delegating the computationally intensive high-accuracy analysis to the cloud server for selected frames.
Data Source
AI summary
According to the present disclosure, a video analysis method and a video analysis system in which video analysis accuracy at a cloud server and an edge is improved is provided. The video analysis method includes a first image analysis step of analyzing an input image frame on the edge side, a difference value estimation step of estimating a difference value between an evaluation value of an analysis result in the first image analysis step and an evaluation value of an analysis result being predicted when the input image frame is analyzed by the cloud server, and a filtering step of determining whether to transmit the input image frame to the cloud server, based on the difference value.