Opt-in Camera: Person Identification in Video via UWB Localization and Its Application to Opt-in Systems

2024-09-30

arXiv.org

AI Summary

To reconcile the tension between privacy preservation and precise identity binding in video surveillance, this paper proposes a privacy-by-design, actively authorized video recording system that continuously records only individuals wearing Ultra-Wideband (UWB) tags and explicitly granting consent. Methodologically, the system fuses UWB-based ranging with multi-object visual tracking and introduces a novel constrained linear optimization algorithm for trajectory matching, enabling accurate cross-modal pedestrian identity association. It further incorporates joint UWB-camera geometric calibration and an Unscented Kalman Filter (UKF) with adaptive hyperparameter tuning to mitigate non-line-of-sight (NLoS) induced localization errors. Experimental evaluation in enclosed environments (8–23 persons) demonstrates near real-time operation at 10 fps, high authorization recognition accuracy, low end-to-end latency, and strong robustness against occlusion and NLoS interference.

Technology Category

• Computer Vision: Motion & Tracking
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Privacy

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Security and Privacy: Large-scale security measurements
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

This paper presents opt-in camera, a concept of privacy-preserving camera systems capable of recording only specific individuals in a crowd who explicitly consent to be recorded. Our system utilizes a mobile wireless communication tag attached to personal belongings as proof of opt-in and as a means of localizing tag carriers in video footage. Specifically, the on-ground positions of the wireless tag are first tracked over time using the unscented Kalman filter (UKF). The tag trajectory is then matched against visual tracking results for pedestrians found in videos to identify the tag carrier. Technically, we devise a dedicated trajectory matching technique based on constrained linear optimization, as well as a novel calibration technique that handles wireless tag-camera calibration and hyperparameter tuning for the UKF, which mitigates the non-line-of-sight (NLoS) issue in wireless localization. We realize the proposed opt-in camera system using ultra-wideband (UWB) devices and an off-the-shelf webcam installed in the environment. Experimental results demonstrate that our system can perform opt-in recording of individuals in near real-time at 10 fps, with reliable identification accuracy for a crowd of 8-23 people in a confined space.

Problem

• Privacy-preserving camera recording only consenting individuals
• Matching UWB tag trajectories with visual pedestrian tracking
• Real-time opt-in system with reliable identification accuracy

Innovation

• UWB tags for opt-in localization
• UKF and constrained linear optimization
• Novel calibration for NLoS mitigation