stop trusting cloud cameras!! (here's what I use instead)

NetworkChuck Technology Careers Career & Money

About this video

The creator replaces cloud-connected cameras with Frigate, a completely local, open-source AI surveillance system, arguing cloud feeds can be exposed or sold. Frigate offers object detection, facial recognition, license-plate recognition, semantic search, and Home Assistant integration while staying off the internet. His quick setup uses a Raspberry Pi running Frigate in Docker (web UI on port 5000) and cheap Reolink E1 Pro cameras he puts at about $56 each, with the hard requirement that any camera support RTSP; ONVIF enables pan/tilt/zoom control. He tests the RTSP stream with FFmpeg, writes the Frigate config, uses the low-quality substream for detection and the main stream for recording, enables person/keyboard tracking, and sets motion-based recording (his config retains seven days). CPU-only detection strains the Pi — about 41% CPU and slow inference with one camera — but a Hailo AI HAT+ drops inference to about 9 milliseconds in his test. Scaling to 10 cameras, he moves to a desktop with an NVIDIA 2070 for decoding and a roughly $100 Google Coral USB accelerator for inference. The system then broke his Wi-Fi roughly every 12 hours: he traced it to Wi-Fi airtime saturation (a 29% TX retry rate on one access point), not bandwidth, plus a known Reolink RTSP degradation issue, and fixed it with constant bitrate/TCP, go2rtc, staggered daily camera reboots, and a third access point.

Key takeaways

Frigate keeps surveillance local: no cloud subscription or internet exposure, with AI detection, face recognition, license-plate recognition, and semantic search on your own hardware
Camera buying rule: it must support RTSP (Reolink E1 Pro used here); use the substream for detection and the main stream for recording to save resources
Install via Docker Compose on almost anything, even a Raspberry Pi, but add an AI accelerator — Hailo HAT+ or Google Coral USB — once detection load grows
Wireless cameras at scale can saturate Wi-Fi airtime rather than bandwidth: watch AP TX retry rates, prefer wired cameras, lock cameras to assigned APs, and use static IPs
For flaky Reolink RTSP streams, the creator’s fixes were constant bitrate (‘fluency first’), TCP instead of UDP, go2rtc to avoid multiple direct camera connections, and staggered scheduled reboots

Episode details: published December 15, 2025 · 39:25 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are NetworkChuck’s own claims, shared for education — they are not typical or guaranteed.

← Previous: I’ll never use n8n the same…… · Next: Ethernet is DEAD?? Mac Studio is 100x FASTER!! →

All NetworkChuck videos · NetworkChuck profile · NetworkChuck on YouTube