Masked Face Recognition by Applying SSD and ResNet Model for Attendance System
Efanntyo, Aditya R. Mitra · 2021
The rapid development of technology in artificial intelligence (AI) includes face recognition. Some of these developments in face recognition systems can be found easily in everyday life, such as entry access, payment, or attendance recording systems. Notably, the demand for image-based face recognition methods to record attendance increases due to its effectiveness and efficiency. Compared to conventional methods like fingerprint or RFID, facial recognition offers better results. On the other hand, with the arising needs in dealing with the novel coronavirus (COVID-19) pandemic, official health protocols require mask-wearing and maintaining a minimum distance of 1 meter between individuals to prevent the spread of the virus. Along with such a situation, a face recognition system to record attendance can reduce the occurrence of direct contact and allow each individual to maintain a safe distance, including from the attendance device. This paper presents the performance of a masked-faced recognition system that implemented SSD (Single Shot Detection) and ResNet feature extraction. The face recognition system application developed using Python and related libraries show a stable level of masked face recognition accuracy. This evaluation was made at predetermined distances between the face and the camera and measured at room lighting of 200 lux with an average accuracy of 67%. The application also has a feature to send notification emails to every employee who is unable to attend work on their scheduled workdays.