Three boxes,
three stories.
On the side, the front and the back of the house hangs a box with a camera inside. The images pass through a model that recognises whether the box is occupied and which species is breeding.
- Occupied
- 1 of 3
- Eggs total
- 11
- Hatched
- 10
Nestkast Achter
Nestkast Midden
Nestkast Voor
Status is updated on every change the system recognises. A logged-in camera grabs an image every few minutes, which is classified by a model as empty or occupied. On a change, a row is written to the database, and that row ends up here.
Each box has a wifi camera with audio. A Raspberry Pi pulls an image via the Tuya cloud every few minutes and passes it to a small neural model (MobileNetV2 with a custom classifier head). The model has been trained on thousands of frames from these three boxes. By day and at night in infrared.
On every change (from empty to occupied, or from building to eggs) a row is written to the database. Counting eggs and chicks is done by hand. No model is good enough to see through twigs.