Dynamic Anonymization of Video Training Data for Person Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video surveillance technologies for monitoring individuals face challenges in balancing privacy concerns with the need for training data, as manual processing of video data for machine learning models raises privacy issues.
Innovation Solution
A method and system for dynamically selecting levels of anonymization in training data feeds, including blurring faces, replacing them with computer-generated images, or using similar-gender faces, to create processed data feeds for training ML models while maintaining privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If video data is provided for training ML models, then training effectiveness is improved, but privacy concerns worsen
Solution Approach 1:
The patent applies local quality by selectively anonymizing only the face regions of video frames while preserving the rest of the visual information. This allows the training data to maintain sufficient detail for effective ML model training while specifically protecting the privacy-sensitive facial features. The anonymization is applied locally to specific regions rather than uniformly across the entire video data.
Solution Approach 2:
The patent implements dynamics by providing multiple levels of anonymization (first level with face blurring, second level with additional anonymization) and allowing the system to dynamically select between them. The central node can request different anonymization levels based on the specific training requirements, creating a dynamic balance between privacy protection and training effectiveness rather than using a static anonymization approach.
2Manufacturing precision
If manual processing of video data is performed, then training data quality is improved, but privacy exposure increases
Solution Approach 1:
The patent applies preliminary action by performing automated anonymization processing on video data before it is transmitted to or used for training ML models. The training data provider automatically applies anonymization filters to obscure facial features in advance, eliminating the need for manual processing that would require human reviewers to view and process unanonymized video data, thereby preventing privacy exposure during the data preparation phase.
Solution Approach 2:
The patent replaces the mechanical system of manual video data processing with an automated computational system. Instead of having human operators manually review and process video frames (which would expose them to privacy risks), the system uses automated image processing algorithms to apply anonymization filters, substituting human mechanical processing with computational processing that can be performed without human exposure to the raw video data.
3Object-affected harmful factors
If high level of anonymisation is applied, then privacy protection is improved, but training data utility deteriorates
Solution Approach 1:
The patent implements dynamics by providing multiple levels of anonymization (first level with face blurring, second level with additional anonymization) and allowing the system to dynamically select between them. The central node can request different anonymization levels based on the specific training requirements, creating a dynamic balance between privacy protection and training effectiveness rather than using a static anonymization approach.
Solution Approach 2:
The patent applies parameter changes by varying the degree of anonymization applied to the video data. Instead of using a fixed high level of anonymization that would completely obscure identifying features, the system adjusts the anonymization parameters (such as the strength of blurring or the extent of pixelation) to achieve the minimum necessary protection while preserving training utility. Different parameter settings can be applied based on the specific training task requirements.
Data Source
Figure 1~4
AI summary
It is provided a method for enabling training of a machine learning, ML, model, for monitoring a person based on a data feed capable of depicting a person. The method is performed by a training data provider (i). The method comprises: obtaining (40) a data feed capable of depicting the person; selecting (42) a level of anonymisation, from a plurality of levels of anonymisation; anonymising (44) the data feed according to the selected level of anonymisation, resulting in a processed data feed; and transmitting (47) the processed data feed as training data for training a central ML model in a central node. Different levels of anonymisation are available and a change in level can be requested by the central node.