Face recognition system for attendance with region based convolutional neural network

In addition to biometric features of people such as eyes or fingerprints, the most distinctive and basic feature that can be used without intervention or contact is the face. The aim of this study is to develop a face recognition system using deep learning technique and convulational neural network (CNN) algorithm. To avhieve this, an original data set including 400 different photos of twelve students was created by the researchers. Photos of the students were taken in their own classroom by changing their places and angles to the camera to create different angles with different sitting patterns. The data set divided into training and test sets, and each students were labeled individually in each photos. After the training process was carried out, face recognition process was testing on the test set. Training and testing of R-CNN models created with AlexNet, GoogleNet and ResNet50 architectures, faster R-CNN, and YOLOv4 models have been completed. To examine the effect of the number of images in the data set on the success of the network; two different training/test ratio; 300/100, and 350/50; were used. The proposed approach yielded promising results in identifying students using face regions. A success rate of 59% was obtained when the 300/100 training/test ratio was used, while the success rate increased to 98% for the 350/50 ratio. As a result, it has been shown that appropriate deep learning models can be used to recognizing the faces easily with a satisfying level. In addition, these models can be used not only for attendance or security purposes, but also for detecting the emotions of the students. Thus, it may be possible to improve the learning environment and evaluate the readiness of students.