First step — optimise the environment. CozZie Glow for light. Heated blanket. Weighted blanket.
Temperature sensors to dial in the tent — cold outside, warm inside, pile the blankets on and let
the sensors confirm the numbers. The old IP cameras and NVR from a decommissioned home
CCTV system came out of storage.

Then the problem: how to get the recorded footage off the NVR remotely.

What followed was a proper investigation. Wireshark to analyse web app commands and see
what the NVR announced on the network. Telnet to attempt remote root access. ONVIF
programs to try standard protocol access. An ONVIF server built on an old Asus gaming laptop
— abandoned when free software recording limits made it impractical. Too much time spent on
a problem that had a slow but simple manual solution.

Download to USB in the morning while you have breakfast. Daily. No backlog. Each new section
of video requires an entirely new download — no selecting multiple days, no cherry picking time
ranges. Once a day. Simple.

The camera module changed everything. !n collaboration with Al, the idea emerged to analyse
the RTSP stream directly from the Pi — monitoring for pixel changes in the video feed. User
defined snapshot intervals. Adjustable sensitivity. Movement intensity rating for each detection.
Timestamped clips. Spike log. All of it running silently overnight while the NVR recorded
continuously as backup.

The first night the sensitivity settings were not quite right. Not sensitive enough. At 2am —
knowing it would not work correctly unless the threshold was dialled down that night — a
reprogram. Because it had to be right. Because incomplete data is worse than no data.
Then sleep.

The system detected quietly. Timestamped. Logged. Rated. Recorded.

127 detections.

127 Detections — Night One

127 clips. Half a morning spent watching them looking for the significant events — or just copy
the raw NVR footage and start over. Too many clips to review efficiently. The pre-buffer feature,
added to capture the moments before each detection triggered, produced glitchy video. An
entire night of buggy clips.

So | watched the raw footage instead. And | timelapsed it.