The face recognition was improved by, every time an authentication is done successfully, the user model is retrained using the captured frame. However some limitations still stand, namely:
the light conditions of the environment affect the success rate of the authentication;
users using glasses have a lower success rate due to light reflection in the lenses;
high quality videos can sometimes be used to bypass the liveliness detection implemented;
Try to implement different algorithms to make the system more efficient