Enhancing Automatic Attendance System using Face Recognition
Alex Bhattarai, Sampada Dhakal, Arun Kumar Timalsina · 2022 IEEE Global Engineering Education Conference (EDUCON) · 2022
Designing an automated class attendance system requires the feature of uniquely detecting each person of the class room. Furthermore, such system should be resistant to varied illumination in the classroom. The recent works on facenet and vggface2 provide promising results in producing high quality facial embeddings as feature vectors. The paper discusses the methodology to identify individual students uniquely. To achieve this, the Support Vector Classifiers are used. The dataset of about 11 classes each with about 1000 faces are used. From the original dataset, two more datasets are created: one with varied contrast and the other with varied contrast and noise. Further, embeddings of these datasets are created and supplied to four proposed models built using Support Vector Classifiers and Neural Network: Tuned SVC with RBF Kernel, SVC with linear kernel, Simple NN and CNN. Although all four models performed similarly on the clean dataset, tuned SVC with RBF kernel is more resilient to noise and variation in contrast. The accuracies of tuned SVC model for dataset-1, dataset-2 and dataset-3 were found to be 0.99334, 0.98990 and 0.90619 respectively.