Frigate
AI-powered NVR with object detection
Frigate is a local NVR (network video recorder) system built for Home Assistant that uses AI-powered object detection on IP cameras. It minimizes resource use and maximizes performance by only looking for objects when and where necessary, performing real-time detection locally using OpenCV and TensorFlow.
TODO: Configure cameras and detection settings in config.yml TODO: Adjust shm-size based on camera count and resolution (calculate: width × height × 1.5 × 20 + 270480 bytes + 40MB) TODO: Consider integration with Home Assistant for enhanced automation TODO: Configure motion detection zones to reduce false positives TODO: Set up recording retention policies in config.yml
GPU override for Frigate — AMD VAAPI video decode for camera streams
Enable by setting GPU_COMPOSE_SUFFIX=amdgpu in config/docker/<hostname>/.env.
The override reuses the shared host GPU_RENDER_GID and GPU_VIDEO_GID settings
defined in config-example/docker/myhost/.env.
Verify these once per host; no service-specific copies are needed.
Also set hwaccel in config/config.yml:
References
- Home: Frigate NVR
- Source: GitHub - blakeblackshear/frigate: NVR with realtime local object detection for IP cameras
- Docs: Introduction | Frigate
- Install: Installation | Frigate
- Config reference: https://docs.frigate.video/configuration/