Abstract:
There has been a great advancement in the field of face recognition and verification in the recent years. Several international conference paper, journals and projects have been published and launched in this regard. Majority of these projects work at a very high efficiency beyond human recognition level. The aim of this project is to develop an efficient face recognition model that can detect and verify the faces of humans through the eyes of CCTV camera footage and efficiently develop a surveillance application based on Deep Neural Networks to notify the user regarding any unwanted activity on their property.
For face recognition we use the approach followed by FaceNet: A Unified Embedding for Face Recognition and Clustering proposed by a team of Google Computer Scientist. The novelty in our solution is the application development portion that will lead to market research and development of application of user notification in case of any unwanted person is found in the video stream of the locality. The procedure will be to draw mappings out of frames in the Euclidian Space and use distances as a measure of similarity. These distances are represented using 128-byte face embedding which provides computational efficiency. The project integrates a Django REST framework, Android Application and this deep learning model with face detection one to bring a complete face recognition prototype that can be used on commercial scale level.
