Gait Recognition and Behavioral Biometrics

Video-based gait research typically detects a person, aligns frames, and derives either silhouette-based features or estimated pose/joint trajectories. A gait energy image averages silhouettes across a walking cycle; newer methods learn an embedding from multiple views or sequences. The CASIA-B benchmark contains 124 people recorded from 11 views under normal walking, bag-carrying and coat-wearing conditions. Such benchmark results are conditional on the dataset’s participants, cameras and tasks, not a guarantee of identification accuracy in streets, crowds or every clothing condition. Angle, occlusion, speed, footwear, health and carried items can all change the signal. Behavioral biometrics is a broader label. Keystroke dynamics may compare dwell and flight times; mouse or touch models may use timing and trajectory. These can be signals for risk assessment or authentication, but they are noisy, can drift, and should not be treated as an infallible identity proof. A device’s motion sensors and a camera also raise different collection and consent issues. GDPR Article 4(14) defines biometric data as data from specific technical processing of physical, physiological or behavioural characteristics that allow or confirm unique identification. Under Article 9, the special-category restriction applies when biometric data is processed for the purpose of uniquely identifying a natural person, subject to the regulation’s exceptions. A health inference from movement may raise other data-protection issues, but is not automatically the same legal category. The EU AI Act’s remote biometric-identification provisions likewise have defined scope and exceptions. Avoid stating that gait is universally covered or universally excluded by all biometric statutes.