Six cameras. One corridor. Face, gait, and skeleton captured in 3D from every person who walks through. Matches existing 2D watchlists on day one while building a multi-modal database no other system can.
01 - Capture
Walk-Through Corridor
3 cross-corridor stereo pairs. One camera per wall, opposing each other. Subjects walk between them. Time-synchronized cameras with zero motion blur. No cooperation required.
CORRIDOR: ENTRY CONTROL POINT
CAPTURING
3x STEREO PAIRS | 6 CAMERAS
SUBJECTS: 0
0s
Stop Time Required
4+
Concurrent Subjects
High
Frame Rate
1,700+
Total Frames Per Subject
02 - Extract
Calibrated 3D Skeleton
Stereo triangulation produces real-world 3D joint positions in meters. Body measurements extracted from calibrated geometry, not estimated from pixels.
TEMPLATE: 3D SKELETON
AZ: 0.0
17 KEYPOINTS | 3 STEREO PAIRS FUSED
SCAN: ACTIVE
6
Viewpoints
±2cm
Measurement Accuracy
~5s
Walk-Through
17
3D Joints Tracked
Measurements include height, shoulder width, hip width, inseam, and arm length — extracted from stereo-triangulated 3D joints in real-world units. Limb length ratios are the most stable biometric after DNA: your skeleton doesn't change with weight, clothing, or age.
03 - 3D vs 2D
What 3D Adds to Every Modality
Existing systems capture a flat image. Tesseract captures calibrated 3D — unlocking modalities that 2D physically cannot extract, and improving the ones it can.
3D vs 2D ACCURACY COMPARISON
FUSED: 3D ADVANTAGE
ESTIMATES FROM PUBLISHED RESEARCH
MULTI-MODAL FUSION
+3–5%
Face (Frame Selection)
+13%
Gait (View Invariant)
NEW
Skeletal Biometric
~96%
Fused (vs 91% 2D)
FACE RECOGNITION
3D doesn't change the face model — the same embedding (ArcFace) runs on both. The advantage is frame selection: 6 cameras produce 1,700+ frames per subject. Tesseract picks the best near-frontal, well-lit frame. A single checkpoint camera gives you whatever angle it gets.
GAIT: THE BIGGEST WIN
2D gait recognition is fragile — it depends on the camera angle matching between enrollment and probe. Change the angle, accuracy collapses. 3D gait uses real joint trajectories in meters, producing a view-invariant signature that matches from any future camera angle.
04 - Disguise Resistance
When Face Fails, Fusion Holds
A mask defeats face recognition — for any system. The question is what happens next. 2D face-only collapses. 3D multi-modal fusion degrades gracefully.
COUNTERMEASURE ANALYSIS
IDENTIFIED
SCENARIO 1 OF 4
3D FUSED vs 2D FACE-ONLY
SKELETAL PERMANENCE
Shoulder width, hip width, limb lengths, and proportional ratios are determined by bone structure. No disguise changes them. A 2D system sees only pixels — Tesseract triangulates real dimensions from calibrated stereo pairs.
GAIT PERSISTENCE
Consciously altering gait is unsustainable — motor patterns reassert within 30–60 seconds. The corridor captures 10+ full gait cycles. Enough to detect both the altered pattern and the underlying natural signature.
05 - Gait Biometric
The 3D Advantage
2D gait: ~75% accuracy, breaks when camera angle changes. 3D gait: ~88% accuracy, view-invariant. The same signature matches regardless of which camera captures the subject next.
GAIT SIGNATURE EXTRACTION
CYCLE: 0
3D JOINT TRAJECTORIES: 14 TRACKED
VIEW-INVARIANT SIGNATURE
06 - Match
Day-One 2D Compatibility
Captures in 3D, matches against existing 2D databases immediately. Same face embedding format as current systems. Gait and skeleton build a new index over time that no other system feeds.
3D CORRIDOR CAPTURE
BEST FRONTAL FRAME SELECTED
Face embedding comparison
EXISTING 2D WATCHLIST
MATCHING...
SOURCE: MUGSHOT / PASSPORT PHOTO
Day 1
Face vs 2D Watchlists
Day N
Gait + Skeleton DB Grows
Same
Embedding Format
+3
New Modalities Over Time
BACKWARDS COMPATIBLE, FORWARD UNIQUE
Face embeddings from Tesseract compare directly against existing 2D mugshot databases (ABIS, NGI, BEWL) using the same cosine similarity pipeline. No format changes, no new infrastructure. But every corridor encounter also enrolls gait, skeleton, and body measurements — modalities that grow a 3D database no other system can query. Someone can change their face. Their gait and skeleton follow them forever.
07 - Landscape
No One Does This
Every deployed biometric system works from 2D images. Tesseract is the first system designed to capture calibrated 3D biometrics passively at scale.
CURRENT STATE
Single 2D Photo
IDEMIA, NEC, Cognitec, Paravision — all capture and match from flat 2D images. Face-only. When the face is obscured, accuracy collapses to near-zero. No gait, no skeleton, no fallback.
RESEARCH ATTEMPTS
Estimated 3D from 2D
IARPA's multi-year BRIAR program (tens of millions across performers including MSU, Kitware, and others) attempts to estimate 3D body models from flat drone footage. Accuracy degrades at distance and off-angles — because estimated 3D from 2D is fundamentally limited by the source data.
TESSERACT
Ground-Truth 3D Capture
Multi-camera stereo triangulation of every person who walks through. Real measurements in meters from calibrated geometry. Not estimated from 2D. Not a single depth sensor. Actual measured 3D from 6 synchronized cameras.
INHERENTLY UNSPOOFABLE
Single-camera systems can be defeated with high-quality masks, printed photos, or deepfake video. Tesseract captures synchronized 3D geometry from 6 cameras on opposite walls — you would need a physical entity that walks with natural human gait and has consistent 3D geometry from every angle simultaneously.
EXISTING LEGAL AUTHORITY
TSA deployment operates under ATSA (49 USC 114) — the same authority used for existing CAT-2 systems. DoD collection under DoDD 8521.01E. Intelligence collection under Title 50 / EO 12333. No new legal authority required.
08 - Integration
Database Architecture
Face embeddings feed directly into existing government databases. 3D modalities create a new biometric index alongside fingerprint, iris, and 2D face.
CAPTURE
Stereo Corridor
3x Stereo Pairs (6 Cameras)
PROCESS
On-Prem or Cloud GPU
Detection + Pose + Triangulation
EXTRACT
Multi-Modal Template
Face + Gait + Skeleton
GOVERNMENT DATABASES
DOD ABIS30M encounters | 13M identities
DHS IDENT / HARTBiometric entry/exit | 300M+
BEWL WATCHLIST213K+ persons of interest
FBI NGI + IC PARTNERSCross-agency interoperability
09 - Connectivity
Deploy Anywhere
Cloud processing over standard internet or Starlink. Edge processing with local GPU. The biometric template is under 5 MB — lightweight enough to batch-upload from anywhere.
DEPLOYMENT TOPOLOGY
STARLINK: CONNECTED
TEMPLATE: <5 MB PER ENROLLMENT
EDGE: OFFLINE CAPABLE
~20
Mbps Total Upstream
Starlink
Compatible
Edge
Offline Capable
1
PC Per Corridor
TIER 1: STANDARD
Airport / Major Base
6 cameras connect to an on-prem PC that streams ~20 Mbps to cloud GPU for processing. Standard internet. Direct database integration.
TIER 2: STARLINK
Forward Base / Embassy
Same pipeline over Starlink. ~20 Mbps upstream fits within Starlink capacity. Identical cloud processing, no degradation.
TIER 3: EDGE
Mobile / Remote
On-prem PC with GPU handles all processing locally. No cloud dependency. Templates queued and batch-uploaded when connectivity is available.
10 - Deployment
Existing Infrastructure
Deploy into existing checkpoint corridors. Same locations, same legal authority, fundamentally richer data.
DHS / TSA
Airport Checkpoints
900M+ annual screenings
440 US airports
2,100+ CAT-2 podiums deployed
$128M current IDEMIA contract
Add 6 cameras to existing CAT-2 corridors. Capture 3D geometry and gait as each passenger walks toward the podium. Face embedding feeds IDENT/HART alongside existing 2D. Gait + skeleton feed a new 3D index.
DEPARTMENT OF DEFENSE
Military Identity Intelligence
$7.6B biometrics market
45K-100K daily ABIS submissions
213K+ BEWL persons of interest
6.3% market CAGR
Ruggedized corridor systems at forward base entry points and detainee processing. Build 3D templates of every person. Starlink-enabled for forward deployment. Match against 2D watchlists immediately.
INTELLIGENCE COMMUNITY
Global Identity at Scale
First operational 3D biometric database
Gait + skeleton matching
Cross-agency interoperability
Title 50 / EO 12333 authority
Every corridor encounter adds to a 3D database. A surveillance camera anywhere captures a side-angle image — Tesseract's 3D template renders from that exact angle for matching. Gait and skeleton match where face recognition fails.
11 - Network Effect
The Database Compounds
Day one: face matching against existing 2D databases. Over time: every corridor encounter enriches a growing multi-modal 3D index that makes every future encounter more likely to match.