A comparative study of conventional and CNN-based implementations of facial recognition on Raspberry-Pi

This paper experimentally compares two face recognition approaches implemented on Raspberry-Pi in the smart-door system. The first approach is based on Local-Binary Patterns Histograms. The second one utilizes convolutional networks and deep learning. The paper describes the implementations and reports the results in terms of recognition accuracy and time. It shows that the CNN based approach runs faster and achieves better recognition accuracy than LBP even on small library sets and limited resources of Raspberry-Pi.