Anonymous Workplace Hazard Detection Through Image Deidentification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing safety risk detection systems face resistance due to privacy concerns, potential employee persecution, compliance with varying privacy regulations, and risk of discrimination based on physical characteristics.
Innovation Solution
An anonymous safety risk detection system that captures and deidentifies individuals using machine learning models, masking personal features while retaining personal protective equipment visibility, and analyzes safety risks in deidentified images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If camera equipment is used to identify safety risk, then safety monitoring capability is improved, but employee privacy is compromised
Solution Approach 1:
The system segments the image processing into distinct stages: capturing full images for safety analysis, then selectively masking only the regions containing personal features (face, hands, body). This segmentation allows the system to retain safety monitoring capability while removing privacy-infringing elements, as the masking process separates personal identification data from safety-relevant data.
Solution Approach 2:
The system extracts and removes personal identifying features from the captured images by applying masks to specific regions (face, hands, body). This extraction process takes out the privacy-sensitive information while preserving the rest of the image for safety risk analysis, thereby resolving the contradiction between safety monitoring and privacy protection.
2Reliability
If camera footage is used for safety analysis, then hazard detection is improved, but employee persecution risk increases
Solution Approach 1:
The system extracts personal identifying information from the images by masking face, hands, and body regions. This extraction eliminates the basis for employee persecution while maintaining the ability to detect safety hazards, as the masked images retain all safety-relevant visual information without containing identifiable personal data.
Solution Approach 2:
The masking process acts as an intermediary between image capture and safety analysis. By applying masks that obscure personal features while preserving safety-relevant information, the system mediates between the need for hazard detection and the risk of employee persecution, allowing safety analysis without enabling identification of specific employees.
3Adaptability or versatility
If camera equipment is installed across multiple locations, then safety coverage is improved, but compliance with varying privacy regulations becomes more difficult
Solution Approach 1:
The masking system serves multiple functions simultaneously: it protects employee privacy, enables compliance with various privacy regulations, and maintains safety monitoring capability. This universal application across different locations and jurisdictions eliminates the need to adjust or reconfigure systems for different privacy laws, as the same masking approach satisfies multiple regulatory requirements.
4Measurement precision
If full image data is retained for analysis, then safety analysis accuracy is improved, but discrimination risk based on physical characteristics increases
Solution Approach 1:
The system extracts and removes personal identifying features (face, hands, body) from the images through masking. This extraction eliminates the basis for discrimination based on physical characteristics while preserving safety-relevant information such as equipment usage, environmental hazards, and safety protocol compliance, thereby maintaining analysis accuracy without enabling discrimination.
Data Source
AI summary
The present invention relates to safety risk detection systems and methods and in particular to identifying possible hazards in working environments. The invention has been developed primarily for use in/with identifying safety risks and hazards in relatively high-risk workplaces such as construction sites and industrial sites and will be described hereinafter with reference to this application. The invention specifically relates to a method for anonymously detecting safety risk at a location, the method comprising the steps of: capturing digital images of the location; determining whether the captured digital images include individuals using a machine learning model; deidentifying individuals in the captured digital images to generate deidentified images; and identifying safety risks in the deidentified images using a safety machine learning model.