Eye See You

We implement a face recognition security system that can detect multiple person in the same frame, achieving high accuracy with 90 % True Positive and less than 2 % False Positive.

Comment

Inspiration

Make MLDA@EEE more AI-savvy! We had dreamed of a computer vision for door access control for a long time. Now it's time to implement it!

What it does

We built a hardware system that could detect (and classify) multiple faces in one frame.

  • This model can detect multiple person in the same frame
  • Door will only open with the right person detected
  • High accuracy with 90 % True Positive and less than 2 % False Positive
  • Can easily tune the model for security or convenience

How we built it

We proposed a hardware system based on Raspberry Pi running on Python to detect (and classify) faces. The Raspberry Pi board was connected to a motor that presses door switch to release access. A LED light band is connected to the board, to indicate the validity of the detected faces.

Challenges we ran into

The computing capacity of a single Raspberry Pi is quite limited, which doesn't support high refresh rate detection. We resolved this issue by restricting the resolution to free the computing power.

Accomplishments that we're proud of

  • This real-time detection model runs on Raspberry Pi
  • High accuracy with 90 % True Positive and less than 2 % False Positive
  • Secure model that can be iterated to finetune further

What's next for Eye See You

We plan to finetune the model to make it fit for more scenarios of various light ambience.

Built With

Share this project:

Updates