★ New · feat/fmv

Full Motion Video, geo-locked to the map

Play geospatially-tagged drone and UAS video, extract its embedded MISB KLV metadata, and watch the live sensor footprint track across the terrain as the footage rolls. Bring your own non-standard DJI clips and convert them into STANAG-compliant Full Motion Video — all in a desktop web UI.

STANAG 4609MISB ST 0601RTSP / UDP / SRT DJI → MISBKLV extractionSensor footprintYOLO-E (optional)

What it does

The video knows where it is. Now your map does too.

MISB-compliant motion imagery carries full sensor pose in its KLV metadata — platform lat/lon, frame-center, corner points, slant range, and field of view. 3D Map Explorer reads it in real time and projects the sensor footprint onto the map frame-by-frame.

Parses and encodes the full MISB ST 0601 UAS Datalink Local Set (96-tag enum) — parser and encoder are exact inverses, sharing one constants source.
Live streams from RTSP / UDP / SRT sources; a background ffmpeg tails the KLV elementary stream and fans packets to a bounded per-stream channel.
Footprint computed from corner points, or derived from sensor Horizontal FOV + Slant Range when absent — reports coverage area and Ground Sample Distance (GSD).
Range-aware HTML5 <video> serving with full HTTP Range / HEAD support for instant scrubbing.

GET /api/fmv/{id}/telemetry

// MISB ST 0601 — sampled telemetry (SSE)
{
  "frame": 1284,
  "sensorLat": 34.0522,
  "sensorLon": -118.2437,
  "frameCenterLat": 34.0498,
  "frameCenterLon": -118.2511,
  "slantRange": 1240.5,
  "sensorHFOV": 60.0
}
→ footprint polygon drawn on map

DJI → STANAG conversion

Standardize consumer drone footage

Most DJI video isn't MISB-tagged. 3D Map Explorer ingests DJI telemetry from multiple sidecars and muxes it into a STANAG .ts file — so your library plays back geo-locked just like military-grade motion imagery.

📄

SRT sidecars

Densest telemetry source — per-second subtitle logs muxed first when present.

🗂️

Flight logs

AirData CSV and DJI Fly .txt / .dat logs parsed into ST 0601 frames.

🖼️

EXIF / XMP

Fallback telemetry extracted via MetadataExtractor when no log exists.

🔁

Preference order

SRT → flight log → EXIF, so the richest source always wins per clip.

🎬

MPEG-TS muxing

STANAG .ts output with PMT rewrites and KLV injection, mirroring dji-to-misb.

📊

Async jobs

Poll /api/fmv/dji-convert/jobs/{jobId} for conversion progress.

Optional AI detection

YOLO-E object detection on every frame

An optional zero-shot sidecar posts frames to /detect and draws bounding boxes over the video overlay. Detections can be georeferenced against the KLV footprint and broadcast as CoT — video objects become tactical tracks. It degrades gracefully — hidden entirely when disabled or the sidecar is down. GPU (NVIDIA, CUDA 11.8+) recommended. The AI/ML stack →

Capability	Endpoint	Notes
Catalog & upload	GET / POST /api/fmv	Multipart clip upload or open stream `{ url }`
Clip playback	GET/HEAD /api/fmv/{id}/file	Range-aware bytes for `<video>`
KLV telemetry	GET /api/fmv/{id}/klv · /telemetry	Full extracted KLV, cached + sampled
Live telemetry	GET /api/fmv/{id}/telemetry/events	SSE stream synced to playback
DJI conversion	POST /api/fmv/dji-convert	Video + sidecar(s) → STANAG `.ts`
AI detection	POST /api/fmv/ai/detect	Frame → bounding boxes (optional sidecar)

Web viewer at /fmv/. Optional AI sidecar default AiDetectionServiceUrl = http://localhost:8003 (separate YOLO-E service). KLV scaling mirrors paretech/klvdata.