Improved Face Recognition from Weighed Face Representations for Deepcam

Recent advances in big data and deep convolutional neural network (CNN) have pushed the performance of face recognition significantly and becoming comparable to that of human being. At this moment, data is more important than algorithm when it comes to the contribution to the performance of face recognition. At the same time, video surveillance is becoming increasingly popular in consumer market thanks to the wide adoption of smartphones. This gives a new way to collect more data. This paper shows a way to integrate face recognition to Deepcam, a peer to peer WiFi security camera. By applying real time face recognition on the real time video stream from Deepcam, this paper shows how to improve the performance of face recognition by weighted combination of the face representations of the faces from consecutive frames.