Automatic Attendance Management System using Image Processing

It is common in education institutes to record attendance which is mostly manual. Significant time and effort are being spent into recording attendance in classrooms which is a hindrance for a flow of teaching. To limit the effort and time taken, we need a smart, effective and authenticated attendance system. Generally biometrics like fingerprint, Iris, hand geometry, face recognition and technologies like RFID are being used for smart attendance system even though each of them have their own demerits. Using RFID in schools is not reliable as it may lead to proxy/false attendance. Whereas, finger print recognition is a reliable one but it does not really ensure the presence of the candidate after the fingerprint is taken. Hence there is a need in the market to build a smart attendance system which is reliable, effective, time and effort saving. Currently, techniques of Image processing mixed with face recognition are often used in several industries for various applications and these processes are done without any human intervention. Hence it would be a great fit for building a smart attendance system. We have several algorithms that help in object and face detection. The work aims to select the most reliable algorithm and build a system that takes attendance for students using face recognition and image processing techniques. The proposed system architecture will include primarily an interface with camera and microcontroller that has enough CPU process power to perform image processing. The work involved evaluation of different algorithms like PCA (Principal Component Analysis), LDA (Linear Discriminant Analysis), ICA (Independent Component Analysis), Fisherfaces to underpin the most suitable one for the application. PCA is an effective, robust algorithm and works better with images without disturbances while can even work on Images with disturbances but fails when matrices are singular. Eigen faces are fast, effective but are sensitive to lightning and position of head. However, the Viola-Jones algorithm uses Haar classifier, integral images, AdaBoost filtering, cascading which makes it the best object detection algorithm for frontal face. The application was developed with the Viola-Jones algorithm running on Raspberry PI with a pair of USB cameras. The system when evaluated in a real time classroom, showed an accuracy of up to 75% in different lighting conditions, head positions and other external factors.