MOVING PERSON IDENTIFICATION IN VIDEO SURVEILLANCE SYSTEMS

• Main Authors: A. Y. Solomatin, I. A. Zikratov, A. S. Lyubert
• Format: Article
• Language: English
• Published: Saint Petersburg National Research University of Information Technologies, Mechanics and Optics (ITMO University) 2014-07-01
• Series: Naučno-tehničeskij Vestnik Informacionnyh Tehnologij, Mehaniki i Optiki
• Subjects: video surveillance; motion detection; time series analysis; periodic components analysis; object classification; face identification; immune computing
• Online Access: http://ntv.ifmo.ru/file/article/10383.pdf
• ISSN: 2226-1494; 2500-0373

The paper deals with an approach for a moving person identifying in video surveillance systems. The proposed solution consists of two successive stages. Selecting of a moving human from all other moving objects in a video stream takes place at the first stage. Human identification based on facial image takes place at the second stage. Detection of a human’s movement is performed via representation of the original video stream in a form of time series. Mathematical apparatus of a singular spectrum is applied for that purpose. The presence of motion is determined by analyzing the periodic components of time series constructed from color and brightness data of the original components of initial video stream. Identification of a person based on his facial image is done through representation of a facial image via two-dimensional matrix with the subsequent application of immune computing mathematical apparatus. Then the binding energy is calculated which shows similarity between the input facial image and faces stored in the training set. The proposed solution for a problem of a moving person’s identifying gives the opportunity to work with low quality video stream having a high level of noise or compression artifacts after encoding. The advantage of the method is implementation simplicity. Unlike traditional methods of computer vision, the proposed method does not require significant computational burden due to simple numerical operations. This method does not require pre-filtering of video images, therefore its performance speed is significantly increased.