Repurposing IP Security Cameras into a Self-Hosted, Local-AI Bird Detector

A homelab tinkerer describes wiring the microphones on three existing RTSP security cameras into BirdNet-Go, a Dockerized audio-recognition tool that identifies birds (and, incidentally, bats and frogs) from their calls in real time. The appeal is that all inference runs locally on modest hardware like a Raspberry Pi — no cloud APIs, no subscriptions, and no audio leaving the network unless the owner opts to share it. A recent addition of Google’s Perch v2 model pushes coverage to roughly 14,800 species, up from about 6,000 in BirdNET 2.4.

The setup leans on familiar self-hosting plumbing: MQTT auto-discovery into Home Assistant, Discord notifications, per-species alert rules, novelty tracking that flags first-time visitors, and optional data sharing to the community BirdWeather platform. Because it reuses cameras already deployed for security, the project adds a genuinely useful second function without new hardware. The author even exposed it to friends over a public domain fronted by Cloudflare.

Beyond the novelty — the system once flagged a passing neighbor’s flatulence as a detection event — the piece is a small case study in local-first, privacy-preserving AI. Keeping the models and audio on-premises sidesteps recurring costs, shifting terms of service, and the risk of a vendor shutting the service down, while turning a passive surveillance rig into an always-on biodiversity logger the household actually checks daily.