That is how the sleep study pipeline was born. Not by design. By necessity. A problem that had
a slow but simple solution.

It was meant to be a wake up call. That is exactly what it was.

The footage had to be uploaded. It was too shocking to ignore.

127 clips. All of them legitimate. All of them a revelation.

It is strange, watching yourself sleep, and feeling bad for the person on the screen.

The file had a face.

Video Retrieval and Processing Pipeline

Each morning: USB stick into the NVR, download the previous night's footage, USB stick into
Marco Pollo — the Lenovo IdeaPad running Windows 11.

The processing pipeline runs as a batch file:

1. Timelapse — ffmpeg compresses the raw AVI footage to 0.05x speed. Eight hours
becomes approximately 24 minutes.

2. Label — drawtext filter burns the subject name and date/timestamp directly onto the
footage.

3. Stitch — multiple clips from the same night are concatenated into a single file using the
concat demuxer.

4. Upload — finished MP4 uploaded to YouTube as unlisted. Link logged.

GPU acceleration via AMD AMF reduces processing time significantly on the Ryzen 5
processor.

The result is a complete nightly record — every movement, every event, every anomaly —
compressed into a reviewable timelapse with forensic timestamps that cannot be altered after
the fact.

The NVR retains the raw footage as the unedited source of truth. The timelapse is the working
document.