W.-h. Yun et al.: Fast Group Verification System for Intelligent Robot Service Contributed Paper Manuscript received October 14, 2007 0098 3063/07/$20.00 © 2007 IEEE 1731 Fast Group Verification System for Intelligent Robot Service Woo-han Yun, DoHyung Kim, and Ho-Sub Yoon Abstract — The intelligent robot service is a promising area and based on many researches such as navigation, hardware control, image processing, and pattern recognition. The intelligent service robot provides the personalized service to a user without the help of a user. In case of the home robot service, the robot providing intelligent services needs a fast verification process which knows whether the user is a family member or non-family member to provide the differentiated services to each group member. In this paper, we develop the fast group verification system for providing intelligent robot services. The proposed system consists of face detection part, preprocessing and feature extraction part, and group verification part. Experimental results show that the proposed system achieves better accuracy and faster computational time than other well known methods and our system is suitable for the intelligent home robot service 1 . Index Terms — Intelligent robot service, group dependent service, PCA, GMM, face detection, face verification. I. INTRODUCTION Recently many research institutes and companies have conducted researches on the intelligent robot. Human-robot interaction is an important and promising part for the intelligent robot service. Nowadays, some researchers introduced many algorithms with respect to human-robot interaction such as face recognition, face tracking, emotion recognition and navigation [1][2][3][4]. When the home service robot serves and communicates with family at home, it is needed that the function which notices whether the user is a family member or a guest (non-family member) and provides the differentiated service to each group member. Similar studies have been performed by other scientists in the biometric area. Sebastien Marcel proposed LDA-based face verification using a symmetric transformation to overcome the shortage of enrollment images [5]. Conrad Sanderson et. al. used local features approach and a Gaussian mixture model for fast and robust authentication [6]. Simon Lucey et. al. used fiducial regions such as eyebrows, eyes, nose, cheeks, and mouth to improve the verification performance [7]. However, they focused only on one-to-one matching and needed much computation time. In this paper, we develop a fast group verification system. Ordinary face verification entails one-to-one matching between a user (claimant) and a client (enrollee). However, 1 This work was supported in part by IT R&D program of MIC & IITA [2005-S-033-03, Embedded Component Technology and Standardization for URC] Woo-han Yun, Do-Hyung Kim, and Ho-Sub Yoon are with Intelligent Robot Research Division, Electronics and Telecommunications Research Institute, Daejeon, Korea (e-mail: yochin, dhkim008, yoonhs@etri.re.kr). the process that we develop focuses on one-to-group matching verification between a user and an enrollees group (clients group, such as family). This group verification process enables the home service robot to know the user is a member of the family or a guest for providing the group-dependent services. The example of group-dependent home service using the proposed system is illustrated in Figure 1. Fig. 1. An example of intelligent home service. The home service robot serves and communicates with a family and guests at home. When a person is coming to the robot, the robot detects the face of the person and knows whether he is a family member or a guest. If the person belongs to a family, the robot gives a briefing on the schedules of the family or tells the day’s telephone messages. On the other hand, if the person is a guest, the robot recommends a differentiated service such as a brief introduction to the house or calls the host on the telephone. These home robot services are based on the group verification system. For these services, the verification system must be fast and reliable. So, in this paper we introduce a rapid and trustworthy verification system for intelligent home robot service. Our proposed system consists of three processing modules: face detection module, preprocessing and feature extraction module using histogram equalization and PCA, and group verification module using modified UBM-GMM. In section 2, we describe the proposed group verification system. We present experimental results and discussions in section 3. We conclude this paper in section 4. II. PROPOSED GROUP VERIFICATION SYSTEM FOR INTELLIGENT ROBOT SERVICE A. System configuration The summarized block diagram of the proposed system is shown in Figure 2. The proposed system is composed of three processing modules. The first module is the face detection part