A Face Recognition System Based on Local Binary Patterns and Support Vector Machine for Home Security Service Robot

In this paper, we present a real-time face recognition system for home security service robot which can be applied to recognize the person's face in front and give a warning when the identity of the person is a stranger. Considering the complexity of the actual situation, there might be some errors causing by the following factors like the angle, the size, the environment and the illumination of the human face, which can hardly be avoided. Our system is designed to reduce and eliminate the influence of the factors above. This article first used Local Binary Patterns (LBP) to detect face and cut out the face region. Then, based on the collected and pretreated face database, the target face could be recognized utilizing Support Vector Machine (SVM). However, the given face database are always irregular in size, position of the face area, scale and some other factors. The proposed face recognition system has been tested in practical application for home security service robot. The results indicate that our face recognition system is effective and real-time. It works well in multi-face recognition and stranger identification, which meet the demand of robot.

I. Introduction

Nowadays, people's security awareness and requirement for home service is gradually rising. How to combine home service and security on one robot is worth studying. On the other hand, face recognition, which is a biometric identification technology based on human face feature information, has been a hot research topic for a long time [1], [2]. The face recognition system proposed in this paper, where the camera was used to capture the video stream, can automatically detect and track the human face in the image frame, and then recognize the human's identity. However, the human faces in the video frame are not always completely front faces. Hence, the method should be able to detect the face with angle. According to some recent researches, the common methods of face detection include harr-like feature [3] and LBP feature [4]. It is found that LBP works better in incomplete frontal human faces. In face recognition, eigenfaces [5] and fisherfaces [6] have been widely used. Moreover, SVM [7], [8] with many unique advantages in solving small sample, nonlinear and high dimensional pattern recognition problems has been applied to handwriting recognition, 3D object recognition, face recognition, text image classification and other practical problems, showing a good learning ability. Compared with eigenfaces and fisherfaces, recognition using SVM has a higher accuracy at Labeled Faces in the Wild (LFW) face database [9]. The decision rules obtained from the limited training samples can still get a small error in the independent test set, which indicates the SVM face recognition is very effective in practical application.