Looking at fixing this issue in future with a 3D stereoscopic camera. This would require a more sophisticated facial recognition solution.
Instead of training a new ML model it may be possible to process the stereoscopic inputs individually, and if they both pass, accept the user as authorized.
Looking at fixing this issue in future with a 3D stereoscopic camera. This would require a more sophisticated facial recognition solution.
Instead of training a new ML model it may be possible to process the stereoscopic inputs individually, and if they both pass, accept the user as authorized.