A classroom check-in algorithm based on face recognition

The problems of traditional classroom clocking-in methods being inefficient and easy to receive human interference are gradually emerging. Thus, we propose a classroom face check-in algorithm based on deep learning. The proposed algorithm consists of four modules: face detection, feature extraction, recognition and attendance. The MTCNN algorithm is used in the face detection algorithm module. It is mainly optimized for the actual classroom environment where the students in the back row have smaller faces, posture transformation, occlusion and other problems. Inception-ResNet-v2 architecture is selected in the face feature extraction module. The original network is restructured to prevent overfitting and reduce redundant computation by removing the classification layer, adding the Dropout layer, and L2 normalization. KNN is used for recognition based on the face features. In the case of attendance statistics, the number of attendees and attendance is calculated as the attendance rate. The proposed algorithm achieves accuracy of 94.4% which is higher than SVM's 88.1%, and performs better in indicators such as F1-score and recall rate. The proposed algorithm is proved by these experimental results to be able to effectively improve the efficiency of attendance and reduce the management burden of teachers.