NoPanopticon

Every person has a distinctive walking pattern—a combination of stride length, cadence, arm swing, posture, and weight distribution that is difficult to consciously change and largely consistent over time. Gait recognition systems analyze this pattern from video footage and convert it into a biometric signature that can be searched against a database of known signatures.

Early systems required specialized depth cameras or pressure-sensitive floors. Modern AI-based systems extract gait signatures from ordinary 2D video footage, including low-resolution or partially obscured images. A person does not need to face the camera; the system can work from the back or side. Processing can happen in near-real time on live camera feeds or retrospectively on archived footage.

There are two main technical approaches: model-based methods that reconstruct a skeleton or silhouette of the body and track joint angles, and appearance-based methods that treat the walking body as a visual pattern and use machine learning to find matches without explicit body modeling. The latter are increasingly dominant because they work well on low-quality footage.

Gait data can be combined with other biometrics—face, height, clothing color—to enable tracking even when any single identifier fails. A camera network that captures a person from multiple angles can build a composite biometric profile that is extremely difficult to evade.

Gait recognition operates covertly and at a distance. A person walking through a city, protest, clinic entrance, or place of worship can be identified and tracked without ever being aware they have been biometrically captured. Unlike facial recognition—which has drawn legal challenges and some bans—gait recognition has almost no legal framework governing its use in the United States.

People with disabilities, injuries, prosthetics, or age-related changes in gait may be misidentified at higher rates. The demographic bias research that has been done on facial recognition has not been systematically replicated for gait, meaning disparate error rates may be significant but undocumented.

Gait recognition makes "covering your face" an ineffective countermeasure. At protests, demonstrations, or any public gathering, individuals who take steps to preserve their anonymity from facial recognition can still be identified and tracked by their gait. This significantly narrows the practical scope of anonymous public participation.

Because gait systems can run on existing camera infrastructure, cities that have invested in camera networks for other stated purposes can add gait surveillance without deploying new hardware—a pattern called "function creep" that allows surveillance capability to expand well beyond what was publicly authorized.

Gait recognition runs on standard IP camera infrastructure; no specialized hardware is required for modern AI-based systems. Higher-resolution cameras and wider fields of view improve accuracy, but even commodity cameras at 720p or 1080p are sufficient for many commercial systems. Specific hardware contexts include:

• Standard IP cameras (Axis, Hikvision, Dahua, Hanwha) – any high-resolution camera can feed footage to a gait analysis backend
• PTZ cameras (pan-tilt-zoom) – can track a subject across a scene once gait identification triggers an alert, following them from camera to camera
• Depth cameras (Intel RealSense, Microsoft Kinect) – used in controlled environments (airports, transit checkpoints) for high-accuracy 3D gait capture
• Millimeter-wave radar – emerging hardware that captures gait signature through walls or clothing without a camera; used in research and some border security contexts
• Wearable sensors / smartphone accelerometers – passive gait data can be extracted from a phone's motion sensors, enabling identification without any camera

• Watrix (China) – one of the first commercial gait recognition companies; software deployed by Chinese police and security forces at scale; can identify individuals from up to 50 meters using standard cameras
• SenseTime (China) – large AI company whose surveillance analytics platform includes gait as part of a multi-biometric tracking suite; under US export restrictions since 2021
• Avigilon (Motorola Solutions) – Appearance Search technology tracks individuals by body shape, clothing, and gait-like movement patterns across camera networks; marketed as not using facial recognition while achieving similar tracking outcomes
• Briefcam (Canon) – video analytics platform used by law enforcement worldwide; includes "person of interest" tracking that uses appearance and motion attributes including gait
• Palantir – data fusion platform capable of ingesting gait and appearance data from multiple camera vendors and correlating with other records
• BioSig-ID – behavioral biometrics company developing gait and movement-based identification
• Academic and government labs (MIT, Carnegie Mellon, DARPA Human ID) – gait recognition was originally developed under DARPA's Human Identification at a Distance (HumanID) program; much foundational research comes from US defense-funded universities

China has the most extensive known deployment of gait recognition at scale, integrated into the country's "Sharp Eyes" national surveillance network alongside facial recognition and license plate readers. Chinese police have publicly confirmed using Watrix gait software to identify individuals in crowds.

In the United States, gait-like appearance tracking is already operational through Avigilon (Motorola Solutions) and Briefcam systems deployed in dozens of cities, though vendors carefully avoid using the term "gait recognition" in marketing. The FBI and DHS have funded gait recognition research, and the technology appears in some airport and border security contexts.

The UK's Metropolitan Police has evaluated gait recognition tools. Several EU member states have tested or deployed the technology at border crossings.

There is currently no US federal law regulating gait recognition. No state has enacted a comprehensive ban. The absence of legal guardrails—combined with its covert nature and compatibility with existing camera infrastructure—makes gait recognition one of the fastest-growing biometric surveillance risks with the least public awareness or oversight.