Study of Class Digital Attendance System Base on Simple Computer and Face Recognition Technique
Education can determine one’s life, and the protection of students’ rights, along with the quality and efficiency of education, has become an important issue. Students should be responsible for their schoolwork and study, and teachers have the obligation to care about students’ learning. In consideration of the effect of education and the concern for students, this study proposes a class attendance system featuring the recognition of students’ faces. Aside from recording the attendance of all students in each lecture, the system can be used for roll call and the observation on students’ learning. It consists of two parts. In the first part, Raspberry, which supports image processing, is used for the front-end processing and placed at the entrance of the classroom; Raspicam and face detection enable the recording of the images of students, store the images in Network File System (NFS client) and send them to the back-end processing. The second part involves the Linux Server back-end processing system and the establishment of the NFS server and the Web server. For the first step, the images of students in the NFS are processed for the recognition and analysis of faces. The algorithm Personal Component Analysis (PCA) is adopted for the training and recognition of faces, and the results of the recognition are added into the database; then, the Web server capture the information in the database and show it for the external users. The face recognition technology is used to recognize the faces of 10 training samples and 3 non-training samples which use the system for several times within a month. The two type of face selection range for the training samples is considered; the threshold values for recognizing those unidentified are set and adjusted; the revised version featuring the addition of new samples for new training is proposed. The training samples with a smaller face selection range can reduce the background and other information irrelevant to face characteristics and thus have a significantly higher accuracy of recognition. According to the results of recognition, the accuracy of the new training-based revised version of the training samples with poor dynamic feedback recognition can be as high as 92%. Currently, there is revised version withonly one extra training sample, and all the training samples are practical in different face ranges.
Keywords— Face Recognition, Class Attendance, OpenCV, Raspberry Pi.