IEEE Second International Conference on Data Stream Mining & Processing August 21-25, 2018, Lviv, Ukraine 978-1-5386-2874-4/18/$31.00 ©2018 IEEE 609 Development of Real-time Face Recognition System Using Local Binary Patterns Maksym Kovalchuk Department of Information Computer Systems and Control Ternopil National Economic University Ternopil, Ukraine kow.max7@gmail.com Vasyl Koval Department of Information Computer Systems and Control Ternopil National Economic University Ternopil, Ukraine vko@tneu.edu.ua Anatoliy Sachenko 1,2 1 Department of Informatics Kazimierz Pulaski University of Technology and Humanities in Radom, Poland 2 Research Institute for Intelligent Computer Systems Ternopil National Economic University sachenkoa@yahoo.com Diana Zahorodnia Research Institute for Intelligent Computer Systems Ternopil National Economic University Ternopil, Ukraine dza@tneu.edu.ua Abstract— This paper describes the development of real- time human recognition system in video streams with the help of Local Binary Patterns (LBPs). The description of the system architecture, face detection process, additional methods for recognition accuracy increase and the method of image comparison based on center-symmetric LBPs are given. Keywords— video analytics; face detection; face recognition; I. INTRODUCTION Video analytics is a widely used technology that uses computer vision techniques to collect various information based on the sequence of frames received from video cameras online or from by using optic flow. This technology can be used in video surveillance, security systems, restricted access systems, pay systems, criminal identification, etc [1]. One of the tasks solved by video analytics is the recognition of faces in video streams. The solution of this problem primarily has a direct application in access control and identification systems [1]. Face recognition is a difficult task to implement due to the variable conditions for visualizing the face such as lighting, the position of the head in relation to the camera, facial expressions and other factors [2-4]. When designing systems, developers try to avoid the negative impact of these factors by imposing severe limitations on the process of acquiring images of the individuals, but the most practical problem is the problem of recognizing faces fast enough to process them with high rate of correct recognitions [2]. In recent years, there has been significant progress in this area, largely due to the improvement of the hardware and computer vision libraries [2]. There are several methods for face detection and face recognition [2-4], which were considered as possible options for implementation. A. Methods of Face Detection: • The Viola-Jones Method; • AdaBoost (Adaptive Boosting); • Support Vector Machine; • Convolutional Neural Network Methods (YOLO); • SNoW (Sparse Network of Winnows). Each of the methods has its main strong points and drawbacks [5]. The Viola-Jones Method: high detection rate due to the use of the cascading classifier, but imposes restrictions on the position of the face upon detection [5]. AdaBoost: high speed of work, but is sensitive to noise and data outliers [5]. SnoW: high speed of work due to sifting the components of the vector of signs, sensitive to noise [5]. Neural Networks: computational complexity and sensitivity, but high detection accuracy with proper network settings [5]. The Viola-Jones method was selected as a method of choice as it includes an improved variation of AdaBoost, has very high detection speed and high detection accuracy of faces in particular compared to the competitors [5], significantly reduces the computation and is famous for face detection with very low false positive rate [5]. B. Methods of Face Recognition • Methods based on pixel brightness values a) Eigenfaces ; b) Fisherfaces; c) Local Binary Patterns . • Methods based on characteristic points a) Feature-based (structural) methods. • Methods based on neural networks. Although neural networks may be preferable in terms of maximum accuracy the use of neural networks does not justify itself in projects where a 3-5% percent difference in recognition accuracy is not critical [6], the processing capabilities of the hardware is limited, and a trade-off between speed and accuracy leans toward speed [6,7]. Each method is designed to solve a specific problem and has its own peculiarities when working under different conditions [6]. In this regard, it is necessary to select such components in the developed system, which will be more universal and will show good results in various tests for identifying and Lviv Polytechnic National University Institutional Repository http://ena.lp.edu.ua