Passion project. I'm not selling anything right now, and honestly I probably never will — I just want to find something real. So everything's free. Use it, share it, whatever.
Note: this post was written by an AI assistant with direct access to the actual working codebase described below. The technical details are notmarketing copy — they reflect the real implementation.
I've spent the last year building an automated anomaly detection pipeline for sky footage. It runs on your video, filters out everything explainable — insects, birds, commercial aircraft — and flags what's left with a score and a reason.
I trained and calibrated it using Farsight's public sky photography methodology (
https://www.farsight.org/FarsightPress/Photographing_UFOs/Photographing_UFOs_Main_Page.html ) — so it expects footage shot the way serious observers actually shoot it.
Here's what it does:
- Background subtraction + centroid tracking — detects moving objects frame by frame
- Bio filter — suppresses insects and birds using wing-beat frequency analysis (FFT), area oscillation, and shape instability
- Plane filter — suppresses commercial aircraft: straight path + constant speed + edge-to-edge crossing
- IR signature — full-spectrum cameras show elevated red channel on anomalous objects
- Angular velocity — distinguishes a close slow bug from a genuinely fast distant object
What it can't explain gets flagged.
What I need:
Raw sky footage shot to Farsight spec or close to it. Full-spectrum or IR-capable camera preferred. 60fps minimum, 120fps ideal — higher frame
rate gives the bio filter better frequency resolution. Static tripod, pointed at sky, long uncut clips, no post-processing.
Doesn't matter if you saw anything. Clean "nothing happened" footage is just as useful for calibration.
What you get back:
An annotated MP4 clip for every flagged detection — a short video of the object with a tracking box, score, and IR reading overlaid, shareable
as-is. A CSV with one row per anomaly: timestamp in/out, speed, area, IR delta, suppression reason. And full JSON with every detection
frame-by-frame if you want to dig into the raw data yourself.
Your footage stays yours. I don't store it, publish it, or share it. You send it, I process it, I send back the results. We delete footage, so you must upload new, if you want more.