A Drone Just Learned to Recognize You by How You Walk

A quadcopter drone hovering against a cloudy sky

TL;DR:

The system: FarSight is a whole-body biometric recognition platform that identifies people from drone footage by fusing facial recognition, gait analysis, and 3D body-shape reconstruction. It’s designed for ranges beyond 1,000 meters and viewing angles above 20 degrees. ^[1][2]
Who built it: Michigan State University, with collaborators at Purdue, UT Austin, and Georgia Tech. Funded by a $12 million, four-year grant from IARPA, the intelligence community’s research arm. ^[3][4]
Why it matters: Covering your face doesn’t beat this. The system’s lead researcher says “a person’s height, build and gait often contain more useful biometric information than their face” at distance. It identifies you by the way you walk. ^[3]
The gap: Facial recognition bans exist in a dozen U.S. cities. Not one of them covers gait analysis or body-shape biometrics. FarSight operates in a regulatory void. ^[1]
What you can do: Awareness is step one. Vary your walking pattern. Support biometric privacy legislation that covers more than just faces. Push for transparency in how these systems get deployed.

Your Face Is the Least of Your Problems

You probably think of drone surveillance as a camera in the sky. Zoom in, snap a face, run it through a database. That’s the old model.

FarSight doesn’t need your face.

Built by researchers at Michigan State University and funded by a $12 million IARPA grant, FarSight is a “whole-body biometric recognition” platform. It processes drone video through six stages: detection, tracking, image restoration, and then three parallel biometric analyses: face, gait, and body shape. The results get fused into a single identity match against a gallery of known subjects. ^[1][2][3]

The system was designed from the ground up for the specific challenges of aerial surveillance: extreme viewing angles (above 20 degrees), long range (beyond 1,000 meters), and images degraded by atmospheric turbulence. Traditional facial recognition falls apart under those conditions. FarSight was built for exactly those conditions. ^[1][2]

Three Ways to Identify You, and You Can’t Hide From All of Them

FarSight’s three biometric modalities work like this:

Facial recognition handles the cases where a clear face is available. FarSight uses AdaFace, a technique that works with low-quality imagery, plus controllable face synthesis to bridge the gap between lab training data and real-world drone footage. Standard fare, but optimized for altitude. ^[2]

Gait analysis is where it gets interesting. The module (called GlobalGait) doesn’t just look at how you swing your legs. It integrates localized movement patterns with broader spatial and temporal walking characteristics. Your stride length, your arm swing, the way your hips rotate, the rhythm of your footfalls. All of it becomes a biometric signature as unique as a fingerprint. ^[1][2]

Body-shape reconstruction is the most unsettling piece. FarSight attempts to extract what the researchers call a person’s “naked 3D body shape”: separating identity-related features from clothing, posture, and movement. Your skeleton. Your proportions. The things that stay the same no matter what you wear. ^[1][5]

Each modality alone is decent. Fused together, performance jumps by over 11% on identification benchmarks. ^[2]

Tested on 1,055 People Who Didn’t Get a Choice

FarSight was evaluated on IARPA’s BRIAR benchmark (Biometric Recognition and Identification at Altitude and Range). The dataset contains over 350,000 images and 1,300 hours of video from 1,055 outdoor subjects. ^[1][2]

The benchmark includes two scenarios: “FaceIncluded,” where the face is visible, and “FaceRestricted,” where faces are occluded, low-resolution, or otherwise unusable. FarSight was specifically designed to perform in that second scenario: the one where you thought you were anonymous. ^[1]

The system was also evaluated in the 2025 NIST RTE Face in Video Evaluation, establishing it as a state-of-the-art solution under real-world conditions. ^[1]

Current operational range sits at 300 to 400 meters for single individuals. The team’s stated goal: 600 to 700 meters for multiple people simultaneously. The BRIAR program’s target spec goes beyond 1,000 meters. ^[3][4]

$12 Million of Intelligence Agency Money

FarSight didn’t come from a university lab running on grants and good intentions. IARPA (the Intelligence Advanced Research Projects Activity) funded it. IARPA is the intelligence community’s version of DARPA. When IARPA funds something, it’s because the intelligence community wants to use it. ^[3][4]

The $12 million, four-year grant went to a team led by three MSU professors: Xiaoming Liu, Arun Ross, and Anil Jain. Jain is one of the most cited biometrics researchers in the world. This isn’t a fringe project. It’s a well-funded, multi-university effort with collaborators at Purdue, the University of Texas at Austin, and Georgia Tech. ^[3][4]

The research paper lists 20 authors. The system was presented at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2024) and published on arXiv. ^[2]

The intended applications, listed in the research: law enforcement, border security, and surveillance. ^[1][2]

The Law Doesn’t Know This Exists

Here’s the problem. Cities like San Francisco, Boston, and Portland banned facial recognition. Illinois’ BIPA created a cause of action for biometric data collection without consent. Europe’s AI Act restricts real-time biometric identification in public spaces.

Every single one of those laws focuses on faces.

Gait analysis? Body-shape reconstruction? None of these categories appear in existing biometric privacy statutes. FarSight operates in a legal gray zone, or more accurately, a legal void. A police department could deploy gait-based identification from a drone tomorrow and no existing city ban would stop them.

Jay Stanley at the ACLU has flagged the issue. The technology’s adaptability to “any high or distant camera, such as those mounted on tall buildings or border-surveillance towers,” he says, implies its routine use on civilian populations. ^[1]

And the researchers themselves frame the intended use cases as “law enforcement, border security and surveillance.” Not “finding missing persons” or “search and rescue.” Surveillance. They put it in the paper. ^[2]

What This Means for Public Anonymity

Think about what it means to attend a protest when a drone overhead can identify you by the way you walk.

People wear masks. They cover their faces. They do this because they understand facial recognition. FarSight makes that precaution irrelevant. Your gait is a biometric identifier you carry with you every time you move. You can’t take it off.

The BRIAR dataset was collected outdoors. The system is designed for crowds. The funding comes from the intelligence community. Connect those dots and you arrive at a surveillance tool that can identify individuals at a protest, a border crossing, or a public park from half a kilometer away, without anyone on the ground knowing they’ve been scanned.

Whole-body biometric recognition turns every outdoor movement into a potential identification event. That’s not a hypothetical: that’s the stated capability.

What You Can Do (Honestly, Not Much, Yet)

Let’s be real: there’s no uBlock Origin for your walking pattern. But there are things worth doing:

Push for broader biometric privacy laws. If your city bans facial recognition, lobby to extend the ban to gait, body shape, and other whole-body biometrics. The ACLU’s model legislation covers “biometric surveillance,” not just faces: push for that language.
Support transparency requirements. Demand that any law enforcement drone deployment disclose what biometric technologies are in use. You have a right to know if you’re being identified by how you walk.
Be aware of the capability. The assumption that covering your face protects your identity is no longer safe. Act accordingly when considering what information your physical presence reveals.
Watch for FOIA disclosures. IARPA-funded technology migrates to operational agencies. File or support FOIA requests targeting CBP, ICE, and DHS drone programs to learn whether whole-body biometrics are already in use.
Follow the regulation fight. Europe’s AI Act may eventually address gait and body biometrics. The U.S. has nothing comparable yet. State-level biometric privacy bills: watch for language that covers “behavioral biometrics” or “gait analysis.”

The Bigger Picture

FarSight isn’t the first drone biometric system. It won’t be the last. But it represents something new: the fusion of multiple biometric modalities into a single platform designed to identify you even when individual methods fail.

Face obscured? There’s gait. Unusual posture? There’s body shape. Too far away for any single method? Fuse them all and the accuracy climbs.

The researchers built a system that gets better precisely when you try to hide. That’s not a bug. That’s the entire point of a $12 million intelligence agency investment.

Facial recognition was the opening act. Whole-body biometrics is the main event. And the laws haven’t even started to catch up.