Head Count: Privacy-Preserving Face-Based Crowd Monitoring

A team of researchers from the University of Twente and Vrije Universiteit Amsterdam has published a paper titled 'Head Count: Privacy-Preserving Face-Based Crowd Monitoring' on arXiv. The system addresses a critical gap in public safety and urban planning: accurately counting and tracking crowd movement over time and across locations without violating individual privacy. Traditional methods that rely on randomized signals from personal devices have become ineffective due to improved privacy protections in modern smartphones.

The proposed 'Head Count' pipeline offers a novel technical solution. When a camera detects a face, it extracts anonymized features and immediately deletes the original image. These features are processed through a fuzzy extractor to create a unique identifier, which is then inserted into a homomorphically encrypted Bloom filter. This combination of cryptography allows the system to perform 'oblivious set membership testing'—essentially checking if a person has been counted before—directly on the encrypted data. The result is a system that can re-identify individuals for accurate longitudinal counts without any party, including the system operator, being able to decrypt and reveal a person's actual identity.

The initial evaluation presented in the paper shows promising results, suggesting the method is both feasible and effective. This approach could enable applications like measuring foot traffic in retail areas, managing crowds at major events, or studying urban mobility patterns, all while maintaining a strong guarantee of biometric privacy. It represents a significant step toward reconciling the utility of widespread visual sensing with the fundamental right to anonymity in public spaces.