Face recognition on Raspberry Pi based on MobileNetV2
Fushuai Wang, Renren Zheng, Penghui Li, Hanni Song, Dongming Du, Jingchao Sun · 2021
At present, face recognition has been widely used, and face recognition in mobile devices has a broad application prospect. We try to detect and extract faces by extracting Haar-like features and then put the face images into the Convolutional Neural Network (CNN) for training and recognition. We improve the MobileNetV2, use ArcFace for face recognition and classification, and test the CNN face classification program on Raspberry Pi 4B. The training in Raspberry Pi is based on the idea of transfer learning. The features extraction part of the network loads the weight file, and only the fully connected classification layers are trained. Finally, the accuracy of train set and test set reach 99.8% and 94.4% respectively after training 100 epochs in Raspberry Pi. When epoch=10, the accuracy rate reaches the predetermined value. We can consider training 10 epochs in Raspberry Pi, and the total time is about 1h33min.