Abstract:Recent works have shown that wearable or implanted devices attached at different locations of the body can generate an identical security key from their independent measurements of the same gait. This has created an opportunity to realize highly secured data exchange to and from critical implanted devices. In this paper, we first demonstrate that vision can be used to easily attack such gait-based key generations; an attacker with a commodity camera can measure the gait from a distance and generate a security key with any target wearable or implanted device faster than other legitimate devices worn at different locations of the subject's body. To counter the attack, we propose a firewall to stop video-based gait measurements to proceed with key generation, but letting measurements from inertial measurement units (IMUs) that are widely used in wearable devices to measure the gait accelerations from the body to proceed. We implement the firewall concept with an IMU-vs-Video binary classifier that combines InceptionTime, an ensemble of deep Convolutional Neural Network (CNN) models for effective feature extraction from gait measurements, to a Generative Adversarial Network (GAN) that can generalize the classifier across subjects. Comprehensive evaluation with a real-world dataset shows that our proposed classifier can perform with an accuracy of 97.82%. Given that an attacker has to fool the classifier for multiple consecutive gait cycles to generate the complete key, the high single-cycle classification accuracy results in an extremely low probability for a video attacker to successfully pair with a target wearable device. More precisely, a video attacker would have one in a billion chance to successfully generate a 128-bit key, which would require the attacker to observe the subject for thousands of years.

GaitGuard: Towards Private Gait in Mixed Reality

Understanding Person Identification Through Gait

GaitPrivacyON: Privacy-Preserving Mobile Gait Biometrics using Unsupervised Learning

Gait-Based Privacy Protection for Smart Wearable Devices

SafeGait

GaitKeeper: An AI-Enabled Mobile Technology to Standardize and Measure Gait Speed

Pedestrian Attribute Editing for Gait Recognition and Anonymization

Human Identification Based on Gait Manifold

PhysioGait: Context-Aware Physiological Context Modeling for Person Re-identification Attack on Wearable Sensing

User gait biometrics in smart ambient applications through wearable accelerometer signals: an analysis of the influence of training setup on recognition accuracy

Classifying Simulated Gait Impairments using Privacy-preserving Explainable Artificial Intelligence and Mobile Phone Videos

GaitSet: Cross-view Gait Recognition through Utilizing Gait as a Deep Set

IRGA: An Intelligent Implicit Real-time Gait Authentication System in Heterogeneous Complex Scenarios

Gait-Guard: Turn-aware Freezing of Gait Detection for Non-intrusive Intervention Systems

SafeGait: Safeguarding Gait-based Key Generation against Vision-based Side Channel Attack Using Generative Adversarial Network.

PGGait: Gait Recognition Based on Millimeter-Wave Radar Spatio-Temporal Sensing of Multidimensional Point Clouds

AGRS: Automated gait recognition system in smart environment

GaitSet: Regarding Gait as a Set for Cross-View Gait Recognition

Vision-based techniques for gait recognition

A privacy-preserving approach to streaming eye-tracking data

Gait Recognition Using Spatio-Temporal Information of 3D Point Cloud via Millimeter Wave Radar