TOWARDS ACCURATE AND RELIABLE AUTOMATED CLASSROOM ATTENDANCE USING FACE RECOGNITION AND CNNS

In real world classroom settings, traditional classroom attendance methods are often time consuming, prone to human errors, and inefficient in terms of accuracy. Different solutions have been introduced from time to time, to make them easy, accurate and automatic. Use of computer vision and face recognition is gaining momentum as it provides automation and increases the precision and effectiveness of attendance system. The research aims to design automated attendance systems capable of working seamlessly in real classroom settings, focusing on increasing the accuracy as well as reliability for the development of automated attendance system. With the goal of providing effective fast automatic classroom attendance, this research focusses on evaluating performance of different Convolutional Neural Network architectures performance, such as VGGNet, ResNet, and DenseNet. For the evaluation of the performance of these Convolutional Neural Networks, we have used parameters like accuracy, precision, recall, and F1 score for a real world classroom scenario with different lighting conditions, poses, and occlusions. Our research shows that the ResNet performs better with respect to speed and accuracy in comparison to VGGNet and DenseNet. This makes ResNet an optimum choice for real time deployment.