Facebell

A button press begins the sequence by triggering an image capture at the door. The image moves through the recognition service and the result is attached to a visitor event that the dashboard can display later. We kept that sequence explicit during development because debugging an incorrect name in the interface required knowing whether the problem came from capture, recognition, storage, or rendering.

The dashboard shows the latest event alongside a history of previous visits. When the service cannot match a face, it still stores the image and timestamp as an unknown visitor, which lets the homeowner review what happened without relying on the recognition result alone.

The recognition service returns a confidence score, so we separated recognized and unknown states and kept the captured image beside the result. Weak matches fell back to an unknown visitor, and the interface displayed a name only when the service returned enough confidence to support it.

We tested Facebell by starting every run at the physical button and following the event through the recognition service. Those runs exposed problems with camera position, the delay between the press and capture, network communication, and the dashboard refresh that were invisible when each component was run separately.

At the 2023 APSC 103 Innovation Project Final Showcase, someone could press the physical doorbell and watch their visit appear in the dashboard after the recognition request finished. We ran that interaction from the assembled prototype with live camera input throughout the showcase.

Facebell won within the ECEi cohort after the judges tried the doorbell and inspected the resulting visitor history, including the unknown state used when the service did not return a reliable match.