Abstract
Taking attendance is a daily process in most of the institutions and it has become a time taking and error prone process considering the existing methods of doing it manually. By observing the everyday situations in larger classrooms, if you have 50 students in a class, it takes at least 10 minutes every day to mark the attendance and if an attendance sheet is passed to the students, there is chance of proxies or students forgetting to sign the sheet. This project aims at solving these two issues by automating this entire process using the Face Recognition Technologies. In the system I propose the instructor can take a picture of the class in his mobile device and the application instantly provides him with all the details of the people in the classroom. The application will display the names of students who are present to the class, the total number of students in the classroom and the total number of students registered to the class.
This application uses Deep Neural Networks to perform face detection, feature extraction and classification. It uses Multi-Task Cascaded Convolutional Neural Networks (MTCNN) [1] for the face detection, FaceNet [2] for the feature extraction, Linear Support Vector Classifier (SVC) for the face classification and Flask Server to host the web application on a local server.
The results show that the application was successfully tested to detect 50 students from a picture with 100% accuracy. It was trained on 50 student images with only 4 images per student. The training time was around 20 minutes and it takes less than 17 seconds to process an uploaded image and display the attendance details.
These results suggest that, if we could deploy this application on cloud and connect it to a database along with an improved front-end, we could facilitate the attendance system in institutions which helps the instructors and students to save time and improve authenticity of the process.