Patent Pending — Deterministic Object Identity

RFML Radio Frequency Motion Layer

Identity captured at source, not inferred from pixels. A deterministic metadata layer for video production that binds physical object identity to pixel-space representation at the moment of filming.

Provisional Patent Filed
Prototype In Development
SMPTE Standards Compatible
LIVE_FUSION_STREAM // DETERMINISTIC_ID
Live Fusion Demo — Sports Broadcast
RFML FUSION MONITOR
LIVE — 01:23:45:18
RAW BROADCAST FEED
ATL 21 MIA
CAM_01 // 24fps
RFML DETERMINISTIC OVERLAY
#7 MARTINEZ0.97
#12 CHEN0.94
#23 WILLIAMS0.96
#9 OKAFOR0.93
#15 PARK0.95
#4 SILVA0.91
TAGS6 / 6
SYNCLTC LOCKED
ANCHORS8 OK
LATENCY12ms
object_id:"PLR_MARTINEZ_7" bbox:{412,218,44,62} conf:0.97
object_id:"PLR_OKAFOR_9" bbox:{688,244,40,58} conf:0.93
object_id:"PLR_WILLIAMS_23" bbox:{540,280,42,60} conf:0.96
object_id:"PLR_PARK_15" bbox:{620,340,38,56} conf:0.95
OBJECTS TRACKED: 6 / 6 ACTIVE SIDECAR: RECORDING — 4,218 FRAMES BRIDGE LATENCY: 12ms
How It Works
01

Tag

Attach RF tags (UWB preferred) to physical objects on set. Each tag carries a unique identifier and provides real-time spatial position via ranging.

02

Localize

Fixed anchors around the set define a world coordinate frame. The localization engine computes 3D positions for each tag at 10–100 Hz with centimeter-level accuracy.

03

Project

The RFML Bridge fuses RF positions with camera pose, lens calibration, and SMPTE timecode to project each object into pixel space — frame by frame, in real time.

04

Output

Per-frame metadata records carry object identity, world-space position, and pixel-space bounding regions. Sidecar files travel with the video from set to screen.

Interactive Playback — Fashion / Commerce
RFML INTERACTIVE PLAYBACK
COMMERCE MODE
JKT_0142 // uwb:0xA17C
■■■■■■■■■□ 0.97
DRS_0088 // uwb:0xB29D
■■■■■■■■□□ 0.93
BAG_0215 // uwb:0xC44E
■■■■■■■■■■ 0.99
SHO_0067 // uwb:0xD81F
■■■■■■■■■□ 0.96
ACC_0033 // uwb:0xE92A
■■■■■■■■□□ 0.91
CLICK ANY OBJECT TO VIEW PRODUCT DATA
Product Detail
TAGGED OBJECTS: 5 IN FRAME IDENTITY: DETERMINISTIC (RF) TC 01:14:22:08 @ 24fps
The Shift
Computer Vision (Inference)
  • Probabilistic — guesses identity from pixels
  • Degrades under occlusion and lighting changes
  • Cannot distinguish identical-looking items
  • Requires retraining for new products
  • No timecode synchronization
  • Not archival or auditable
vs
RFML (Deterministic)
  • Deterministic — captures identity at source
  • Works through occlusion and any lighting
  • Distinguishes any tagged object by unique ID
  • No retraining — identity is a physical signal
  • Frame-accurate SMPTE timecode sync
  • Full pipeline preserved for audit and re-projection
Metadata Schema (Example)
sidecar.jsonl
{
  "timecode": "01:02:03:12",
  "frame": 110592,
  "fps": 24.0,
  "camera": {
    "serial": "CAM_01",
    "pose_world": { "x": 1.234, "y": 0.456, "z": 1.789 },
    "focal_mm": 35.0
  },
  "objects": [{
    "object_id": "JKT_0142",
    "tag_id": "uwb:0xA17C9B2E",
    "world_pose": { "x": 3.12, "y": 1.07, "z": 0.92 },
    "pixel_bbox": { "x": 812, "y": 214, "w": 188, "h": 402 },
    "confidence": 0.91,
    "sku_refs": ["SKU_881223"]
  }]
}  // one record per frame — world-space + pixel-space per object
Applications

Interactive Commerce

Viewers select objects to purchase. Identity is exact — SKU-level, not category-level.

Continuity & Clearance

Automated tracking of which objects appear in which shots across an entire production.

Production Analytics

Object-level engagement metrics, conversion funnels, and temporal trends.

Rights Management

Per-object licensing constraints evaluated at playback — by territory, time window, or channel.

#12

Sports & Live Broadcast

Deterministic player and gear identification for real-time overlays and statistics.

OBJ_07

AR & Extended Reality

Object identity and pose streams enable AR overlays on companion devices.

Get in Touch

RFML is in active development. If you work in production technology, interactive video, or shoppable media, I'd like to hear from you.

↓ Download Overview (PDF) Contact → contact@rfml.io