Privacy-Preserving Gait Identification via Center-of-Gravity Oscillation Analysis on a Floor-Based Sensing Platform

Gait-based individual identification has attracted significant attention for its unobtrusive and user-friendly nature; however, many existing approaches rely on image-based sensing, raising serious privacy concerns. To address this problem, we propose a privacy-preserving gait identification system that utilizes a floor-based sensing platform measuring center-of-gravity (CoG) oscillations and weight data during natural walking. The system is designed for simplicity and robustness, consisting of eight load cells, an aluminum frame and plate, A/D converters, voltage modules, and microcomputers. Time-series data are segmented according to the Rancho Observational Gait Analysis System, and both time-domain and frequency-domain features are extracted for machine learning-based classification. Experiments with 31 participants under natural walking conditions, including carrying luggage, demonstrated an F1-score of 0.918 through stratified five-fold cross-validation. Even without weight information, using only CoG oscillations, the system achieved an F1-score of 0.817, confirming that distinctive gait signatures are embedded in body sway patterns. Notably, individual differences were pronounced during the foot transition phase. As the system operates solely on low-dimensional, abstract CoG and weight data without capturing identifiable features such as facial images or silhouettes, it offers strong privacy protection. Furthermore, it is robust against external variations including lighting, clothing, hairstyle, and personal belongings. These attributes make the proposed system promising for applications in security, healthcare, and marketing, enabling unobtrusive, privacy-preserving identification based on natural walking behavior.