Continuous Real-time Automated Attendance System using Robust C2D-CNN
Christopher Chun Ki Chan, Chih‐Cheng Chen · 2020
In this paper, we implement a hybrid real-time continuous face recognition automated attendance system (AAS) to capture attendance duration and attendance from CCTV footage - a missing temporal aspect of which many AAS currently lack. Our system consists of three parts. First, we extract additional features via an implementation of 3D facial reconstruction of which it learns detailed facial features from a single image and obtain an image set for additional complementary features via shape aggregation. Second, we apply MTCNN face detector to automatically detect people who enter a room. Third, we apply C2D-CNN face recognition, which combines features learned from original pixels with image representation and decision level fusion which results in a significantly more robust and accurate face recognition system suitable for CCTV H264 footage. The performance of our model is able to detect and recognize students accurately even with partial faces and low-quality images.