BirdCam v2 — On-device Avian Classification

Why a rebuild

The first BirdCam streamed frames to a laptop for classification. That worked, but it tethered the camera to my desk. Version 2 moves inference onto the device so it can run untethered in the field.

What changed

• Swapped the Raspberry Pi for a Jetson for faster on-device inference.
• Added a motion gate so the model only runs when something moves.
• Logged every detection with a timestamp and confidence score.

Early result: ~12 species reliably distinguished at the feeder, with detections written straight to a local database.

Next up: a solar enclosure so it can live outside for a full season.