Edge Video Analysis Model Update via Cloud Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Edge devices in video analysis systems using AI technology for security applications often generate false alarms due to limited resources and uniform hardware, leading to inconsistent analysis quality across different sites.
Innovation Solution
A video analysis system utilizing edge computing that includes a network of edge devices and a cloud server, where edge devices can update their deep learning models based on feedback from erroneously generated security events, and the cloud server selects and transmits optimal deep learning models to improve analysis quality, leveraging external resources and environment-specific models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If edge devices use uniform hardware with limited resources for video analysis, then device complexity is reduced and ease of manufacture is improved, but analysis quality becomes inconsistent and false alarms increase across different sites
Solution Approach 1:
The system dynamically updates deep learning models on edge devices based on environmental feedback. The cloud server collects analysis results and false alarm data from various edge devices, identifies optimal models for specific environments, and pushes updated models to edge devices. This dynamic model updating allows uniform hardware to achieve adaptive performance improvement without changing the physical device configuration.
Solution Approach 2:
The system changes the software parameter (deep learning model version) rather than hardware parameters. By updating the deep learning model parameters on edge devices based on environmental conditions and performance feedback, the system achieves improved analysis quality and reduced false alarms while maintaining the same uniform hardware configuration across all edge devices.
2Productivity
If edge devices continuously process videos with limited resources, then productivity is maintained, but analysis quality deteriorates due to resource constraints
Solution Approach 1:
The cloud server performs preliminary analysis of false alarm cases and identifies optimal deep learning models before pushing them to edge devices. By pre-processing the model optimization work on the cloud server with abundant resources, the system prepares improved models in advance that can be deployed to edge devices without requiring them to have sufficient computational resources for real-time model optimization.
Solution Approach 2:
The cloud server acts as an intermediary between model development and edge device deployment. It collects performance data from edge devices, trains and optimizes deep learning models using abundant cloud resources, and distributes the optimized models back to edge devices. This intermediary role allows edge devices to maintain high productivity with limited resources while achieving improved measurement precision through cloud-based model optimization.
3Device complexity
If the same deep learning model is deployed to all edge devices, then device complexity is reduced, but adaptability to different environments deteriorates
Solution Approach 1:
The system segments the deep learning model deployment by environment type. The cloud server categorizes different installation environments (e.g., indoor, outdoor, different lighting conditions) and assigns specific optimized models to edge devices based on their environment. This segmentation allows the system to maintain simple uniform hardware while achieving environment-specific adaptability through targeted model deployment.
Solution Approach 2:
The system implements a feedback loop where edge devices report analysis results and false alarm data to the cloud server. The cloud server uses this feedback to identify environmental patterns and optimize models specifically for each environment type. The optimized models are then pushed back to edge devices in corresponding environments, creating a continuous improvement cycle that enhances adaptability while maintaining device simplicity.
Data Source
AI summary
A video analysis system is a system in which a video is analyzed through an edge device installed in each site to generate a security event, and when the edge device continuously generates erroneous security events, the video is analyzed through another edge device or a cloud server in parallel, and then a deep learning model enabling the edge device not to generate an erroneous security event is found to change the edge device, thereby maintaining high video analysis quality.