A hands-on walkthrough of migrating from ZoneMinder to Frigate NVR for a home security camera setup. The author runs Frigate on an Ubuntu VM in Proxmox, passes through two Coral AI M.2 accelerators for object detection, and configures six cameras with varying resolutions and frame rates. Key lessons include Frigate's mandatory detection requirement per camera, the CPU cost of high-resolution ffmpeg decoding, and the practical limits of Coral AI at 1440p. The final config runs door cameras at 1440p/5fps for person and cat detection, wide-field cameras on D1 sub-streams for car/person, and a 4K varifocal on a 720p sub-stream. 3D printer cameras bypass Frigate entirely via Home Assistant. The result successfully tracks the author's cat and filters out false positives like bugs and tree shadows better than ZoneMinder.
Table of contents
The Old System ⌗Setting Up the VM ⌗Setting up Frigate ⌗Cat Detection ⌗Lovelace View ⌗Final Thoughts ⌗Questions this post answers
Can Frigate NVR add a camera without enabling object detection?
No, Frigate requires a detect behavior for every camera — you cannot add a camera without object detection enabled. This means cameras you only want to view or record, such as 3D printer monitors, need to be handled outside Frigate, for example by adding them directly to Home Assistant for restreaming. Home lab builders navigating Frigate's camera requirements share workarounds on daily.dev.
How many cameras can two Coral AI M.2 accelerators handle in Frigate at 1440p?
Two Coral AI M.2 accelerators struggle to run four cameras at 1440p/15fps simultaneously in Frigate. A more sustainable configuration is 5fps analysis per camera, using sub-streams (D1 or 720p) for detection on wide-field cameras while reserving the high-resolution main stream for recording only, reducing both Coral and CPU load significantly. Developers tuning Frigate and Coral AI setups find real-world configs discussed on daily.dev.