Facial Recognition

Mission

▹ Create a facial recognition application from scratch using neural networks.

▹ Client was using traditional paper sign-in sheets creating difficulties surrounding long-term attendance tracking and data upkeep.

Technologies

▹ Libraries: NumPy, Pandas, OpenCV, Kivy (Python), TensorFlow

▹ Techniques: Siamese Neural Networks,

▹ Skills: API routing, Database Management, Automation, System Design

Features

▹ Sign-in/Sign-out page for workers

▹ Real-time facial recognition support

▹ Simplistic UI

Outcome

▹ The application had around a 75% success rate when verifying existing individuals. Issues arose for the one-shot verification as the model was prone to overfitting to training data despite data augmentation and cleansing.

▹ Results were improved by storing initial images of target individuals and assigning them to their corresponding usernames and using additional data augmentation techniques.