Dynamic Edge Model Refinement for Video Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video monitoring systems employ pre-built ML models that are not tailored for specific video capture devices or environments, leading to reduced accuracy due to false positives and false negatives in object detection.

Innovation Solution

A system for dynamically refining ML models by using a management service that includes a cloud model and edge devices to analyze video data, selecting training images based on model comparisons, annotating images, and updating edge models through transfer learning to improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If pre-built ML models are used for object detection, then system implementation is simplified, but detection accuracy deteriorates due to false positives and negatives

Engineering Contradiction:
Improvesystem implementation simplicityVSAvoidobject detection accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The system performs preliminary actions by automatically selecting training images and generating annotations before model retraining. The image selection component identifies suitable training images based on comparison between cloud model and edge model detections, and the annotation component automatically generates training annotations, preparing data in advance for model improvement without requiring manual intervention.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements self-service through automated model refinement. The edge device autonomously selects training images, generates annotations, retrains models, and validates improvements without external intervention. The system serves itself by continuously improving its own detection accuracy through automated feedback loops between detection results and model retraining.

Inventive Principle:
Principle #25Self-service

2Loss of time

If pre-built ML models are used, then deployment time is reduced, but model accuracy deteriorates due to lack of environmental adaptation

Engineering Contradiction:
Improvedeployment timeVSAvoidmodel accuracy
Core Design Contradiction:
Loss of timeVSReliability

Solution Approach 1:

The system implements dynamic model adaptation where the edge model evolves over time based on environmental feedback. The model refinement process dynamically adjusts the edge model parameters by retraining with selected training images and annotations, allowing the system to adapt to changing environmental conditions while maintaining rapid deployment capabilities.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system establishes feedback loops where detection results from both cloud and edge models inform subsequent model improvements. The image selection component uses comparison between cloud model and edge model detections to identify training images, and the automated annotation process feeds back improved annotations to retrain the edge model, creating continuous improvement cycles.

Inventive Principle:
Principle #23Feedback

3Measurement precision

If manual model retraining is performed, then model accuracy improves, but processing time and computational resources increase

Engineering Contradiction:
Improvedetection accuracyVSAvoidmodel refinement time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system extracts only the essential components needed for model refinement: selecting specific training images based on comparison results and extracting key annotation information. Rather than processing all available data, the image selection component identifies and extracts only those images most beneficial for improving detection accuracy in specific environments, reducing overall processing requirements.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system changes parameters by automatically adjusting model architecture and training parameters based on environmental conditions. The automated annotation component generates annotations with optimized parameters, and the model retraining process adapts learning rates, batch sizes, and other training parameters to reduce computational overhead while maintaining accuracy improvements.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4105893B1Dynamic artifical intelligence camera model update
Publication Date: 2025.06.25 SENSORMATIC ELECTRONICS CORP
  • EP4105893B1 patent drawingFigure 1
  • EP4105893B1 patent drawingFigure 2
  • EP4105893B1 patent drawingFigure 3

AI summary

A system may be configured to dynamically update deployed machine learning models. In some aspects, the system may receive sampled video information, generate first object detection information based on a cloud model and the sampled video information, and generate second object detection information based on a first edge model and the sampled video information. Further, the system may select, based on the first object detection information and the second object detection information, a plurality of training images from the sampled video information, detect motion information corresponding to motion of one or more detected objects within the plurality of training images, generate a plurality of annotated images based at least in part on the first object detection information and the motion information, and generate a second edge model based upon training the first edge model using the plurality of annotated images.