Initial Recon Flight · 2026-09-09 13:24:10 · Edeby Gård
Drone Pilot & AI Engineer · Jesper Karlsson
| PARAM | VALUE |
|---|---|
| LAT | — |
| LON | — |
| ALT_REL | — |
| YAW | — |
| STATE | — |
| MODE | REPLAY |
| TRACK | 0 pts |
| FRAMES | 0 |
| KEY | VALUE |
|---|---|
| NAME | Initial Recon Flight |
| ID | Edeby-watch.003 |
| TYPE | MANUAL |
| DUR | 486 s |
| SRC | DJI_Flight_Data.srt |
| T | EVENT |
|---|---|
| 03:27 | Health Check Close Up |
| 03:52 | Geofencing Detected |
| KEY | VALUE |
|---|---|
| AC | DJI Mini 3 Pro |
| ISO | 110 |
| SHUTTER | 1/60.0 |
| APERTURE | f/1.7 |
| CT | 5124 K |
| FOCAL | 240 (raw) |
| ALT0 | 5 m rel |
| T | SRC | EVENT |
|---|---|---|
| — | ||
| CLASS | EXP | SEEN | Δ |
|---|---|---|---|
| Chestnut | 12 | 0 | — |
| Pinto | 2 | 0 | — |
| Dark | 9 | 0 | — |
| White | 3 | 0 | — |
| TOTAL VIS | 26 | 0 | — |
EXP = farm register (26). SEEN = verified unique this flight (räknemanus) — not raw Grounding DINO. Dark = svart + brun. Dedup frames add 0. Report visibility + occlusion — never a false-precise total.
Grounding DINO · prompt: "horse." · zero-shot · nadir · no site fine-tune
Map dots = projection probe (pixel→lat/lon via Mini 3 Pro FOV + SRT; flat ground, estimated heading). Not survey-grade. Raw boxes ≠ herd total — SEEN is curated unique (räknemanus).
| ID | T | CONF | LAT | LON | FR | CLR |
|---|---|---|---|---|---|---|
| — | ||||||
| PARAM | VALUE | UNIT | SRC | NOTE |
|---|---|---|---|---|
| WIND | — | m/s | 97170 | |
| CLOUD | — | % | 97170 | |
| SUN_ELEV | — | deg | capture | OK @ capture |
| VIS / RH | — / — | km / % | 97170 | |
| FOG/MIST | — | 97170 | ||
| PRECIP | — | mm | 97170 | |
| TEMP | — | °C | 97170 |
Mist/haze inferred from visibility and humidity. Fog = SMHI symbol 7 or visibility under 1 km.
This is an AI-assisted horse-monitoring application. For this proof of concept, the drone survey is flown manually but is prepared to run autonomously via PX4. The drone collects imagery over the pastures, after which the footage is analysed using the zero-shot Grounding DINO model, visually cross-checked, and fed back as training data. Detections are tallied to produce a head count and keep stable staff informed.
Weather data is provided by SMHI and added as training dimensions, while simulation and future autonomous-flight simulation are supported by self-hosted open source apps running on Oracle Cloud.