로봇환경에서 3차원 콘볼루션 신경회로망을 이용한 비디오기반 얼굴인식

Abstract

A HRI(Human-Robot Interaction) is a system for robot to interact with human by channels of communication. Natural interaction between robot and human is needed to provide human services. One of the HRI's components is a vision of which role is eye in human. So, a face detection and a face recognition are essential. There are convention techniques for face recognition such as PCA(Principal Component Analysis), LDA(Linear Discriminant Analysis) and 2D-CNN(Convolutional Neural Network). These methods are normally process images one by one. In this paper, we experiment video-based face recognition using 3D CNN on the ETRI face video. Training data is 50pictures captured from 1m away, and checking data are 50, 200 pictures captured from 1m, 2m away and The rates of true recognition are 100% and 88% as maximum by varying the number of feature maps.