Future Work

If given more time, we would explore ways to improve our facial recognition algorithm since prediction accuracy is not as good as it can be. The current Haar Cascade would benefit from training on a larger dataset, but in doing so would require a longer train time. Some solutions include running the training process in the background and letting the dispenser continue on the old model for the time being or exploring the use of a web database that processes the photos and feeds the model back to the Raspberry Pi. We would also consider alternative classification models with greater accuracy or potentially implement a dynamic facial recognition model that would improve the user experience and computation time when adding people to the library.

Additionally, we would add emotion-based snack dispensing such that if a person looks sad, they would get more snacks or override the unrecognized classification to receive a good snack as opposed to marbles. Furthermore, the project could be expanded to include more interactive GUI animations, for example while snacks are dispensing, and to have the drawer automatically open and close as opposed to the user needing to open and close the drawer.