Author:
S. I. Ranjitha ,E. Deepankumar
Abstract
One of the main goals for face recognition in surveillance settings is recognising a person recorded on camera or in an image. This means that faces in still photos and video clips must match. High-quality still picture automatic face identification can perform satisfactorily, while video-based face recognition is difficult to do at a comparable level. Video sequences have a number of drawbacks over still image face recognition. First off, the majority of the time, CCTV cameras provides low-quality photos. There is more background noise, and moving objects or out-of-focus subjects might cause photos to become blurry. Second, video sequences often have lesser picture resolution. The resolution of the real facial image may be as low as 64 by 64 pixels if the subject is extremely far away from the camera. Lastly, in video sequences, differences in facial image, including lighting, emotion, position, occlusion, and motion, are more significant. By creating many "bridges" to link the still image and video frames, the method may effectively handle the uneven distributions between still photos and films. In order to match photos with videos and identify unknown matches, the Grassmann algorithm may be used in this research to develop a still-to-video matching strategy with RFID technologies. Matching feature vectors based on deep learning techniques and reading the feature vectors using the Grassmann method. Additionally, when an unknown face is detected, send out an SMS and email notice. After that, provide reports for the attendance system.