It is an Attendance system based on deep learning
Using the viola jones algorithm and the haar cascade features the faces are detection in a video feed
The model is trained on 200 subjects with intially 14 images per subject. Later on various Data Augmentation Techniques were used as:
The images were flipped to get the mirror images which increased the data twice fold
The system is made illumination invariant so as to make it work in various light conditions
The image is cropped so as to contain majority of facial data as during detection only faces are detected which in turn falicitates the recognition
Recognition is done with multi-layer Convolutional Neural Networks trained from scratch. Giving 89% test accuracy and 92.19% validation accuracy. Alongside Overfitting was avoided with the above mentioned Data Augmentation Techniques.