Walking Through Handoffs: Orientation-Robust Gait Authentication for Shared Devices

Shared mobile devices pervade collaborative work, from nurses passing tablets on ward rounds to researchers exchanging phones in the field and staff rotating through shared terminals, yet each handover demands re-authentication that fragments workflow. Gait-based implicit authentication could eliminate this friction by re-verifying users transparently as they walk, using smartphone accelerometers already at hand. Practical deployment, however, has stalled on a persistent obstacle, device-orientation sensitivity. When a phone settles differently in a pocket, raw accelerometer readings shift and recognition collapses. We resolve this barrier through rotation-invariant (RI) feature engineering, extracting six scalar features mathematically immune to any coordinate-frame rotation. Under a dual-condition protocol simulating realistic orientation variability (±45°), conventional methods collapse from 89% to 45% accuracy while our approach maintains 95.85%, a 49-percentage-point advantage. An ablation study reveals that RI features, not model architecture, drive this robustness, with a simple 1D-CNN achieving 95% when given RI input and matching our full system. For CSCW practitioners, a single preprocessing change can make existing gait-authentication pipelines robust to the orientation shifts intrinsic to device sharing.