Selective Anonymization of Visual Streaming Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in selectively anonymizing visual streaming data while ensuring privacy preservation and compliance with regulations like GDPR, especially in scenarios where real-time analytics and data storage are required.
Innovation Solution
A computer-implemented method and system that utilize edge devices to capture and process visual streaming data, identifying anonymizable objects and calculating quantized identities. These identities are then encrypted and sent to a central server for anonymization decisions, allowing for selective anonymization based on white- or black-lists without transmitting sensitive data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If video data is transmitted over the cloud for analysis and storage, then centralized processing and storage capabilities are improved, but data security and privacy protection deteriorate due to vulnerability to attacks and unauthorized access
Solution Approach 1:
The patent extracts and processes personally identifiable information (PII) from video streams before transmission to the cloud. Edge devices perform object detection, facial recognition, and license plate recognition to identify PII, then redact or anonymize this information locally. Only the processed video data without PII is transmitted to centralized servers, maintaining processing power while eliminating security risks associated with transmitting sensitive data.
2Reliability
If anonymization is performed on all visual data to protect privacy, then privacy protection is improved, but analytical value and service personalization deteriorate due to loss of identifiable information
Solution Approach 1:
The patent applies different quality levels of anonymization to different regions and types of data within the video stream. PII such as faces and license plates receive high-level anonymization (blurring, pixelation), while non-sensitive regions maintain original quality. This localized approach ensures privacy protection for identifiable information while preserving analytical value in non-sensitive areas for tasks like crowd flow analysis or incident detection.
3Adaptability or versatility
If edge devices perform object detection and facial recognition to identify PII, then selective anonymization capability is improved, but computational load and energy consumption at the edge deteriorate
Solution Approach 1:
The patent implements a tiered detection approach where edge devices perform lightweight object detection to identify potential PII regions, then apply full facial recognition and license plate recognition algorithms only to those specific regions of interest. This partial action approach maintains selective anonymization capability by accurately identifying PII while reducing overall computational load compared to processing entire video frames with heavy algorithms.
4Reliability
If consent management systems are implemented to comply with GDPR, then legal compliance is improved, but system complexity and operational overhead deteriorate due to frequent verification procedures
Solution Approach 1:
The patent performs anonymization of PII automatically during video processing before any transmission or storage occurs, eliminating the need for continuous consent verification. By proactively removing identifiable information at the edge device level, the system ensures GDPR compliance by default without requiring complex consent management infrastructure or frequent verification procedures during operation.
Data Source
AI summary
The application is directed at a method and system for selective anonymization, wherein the method comprises the steps of capturing visual streaming data, identifying an anonymizable object in the visual data, for which a quantized identity (y) and an individual private key (n) is determined. Based on the individual private key (n) and the quantized identity (y), the first set of encryptions (E1) is calculated, comprising at least two distinct encryptions of the quantized identity. The first set of encryptions (E1) of the quantized identity (y) is sent to a central server, which, in return, sends an exception information indicating if an exception list of the central server comprises a set of exception encryptions (E2) which corresponds to the first set of encryptions (E1). The anonymizable object is then selectively anonymized in the streaming visual data depending on the exception information and an operating mode of the edge device, thereby generating selectively modified visual streaming data and the selectively modified visual streaming data is transmitted to a remote database.