The gaze direction algorithm is not robust in the following ways :
- Detects eyes irregularly with low lighting/profile faces. (Can be solved with adding more classifiers with flags, will increase code length. Any other suggestions?)
- Many eyes detected over one eye. Creating mutually exclusive blocks can be an option.
- Detecting profile eyes. New classifier needs to be developed? Or use one from the pipeline if exists?
The gaze direction algorithm is not robust in the following ways :