Hi,
I’ve completed integration of SafeHaven events and state detection with Frigate (stock NVR/UI). Since Frigate does not yet provide native Metis integration, I implemented a connector that publishes semantic events through Frigate’s Create Event API.
I also validated an experimental path to replace Frigate’s Coral detector flow with Metis. The prototype works, but it still needs hardening and packaging before production use. For now, I’m using the custom event integration as the primary path and plan to revisit the detector integration as a follow-up.
Current progress:
- Semantic state events are integrated into Frigate timeline.
- Left-open workflows are implemented with configurable thresholds.
- Correlation logic is in progress to check for nearby user activity (for example, package pickup or outdoor activity) before escalating alerts.
- Models currently cover latch state, door position, and backyard door scenarios.
- Garage door model training is next.
- Repository cleanup and production hardening are in progress: SafeHaven_v2
Next steps:
- Complete garage door model training and evaluation.
- Finish scenario testing and evidence capture across all workflows.
- Decide on adding face recognition as an optional module after core state workflows are finalized.
Regards,
Shashi
