APPLICATION OF BINARY DESCRIPTORS TO MULTIPLE FACE TRACKING IN VIDEO SURVEILLANCE SYSTEMS

• Published in: Nauchno-tekhnicheskiĭ vestnik informat͡s︡ionnykh tekhnologiĭ, mekhaniki i optiki Vol. 16; no. 4; p. 670
• Main Author: Oleinik, A L
• Format: Journal Article
• Language: Russian
• Published: Saint Petersburg St. Petersburg National Research University of Information Technologies, Mechanics and Optics 01.07.2016
• Subjects: Algorithms; Cameras; Closed circuit television; Face recognition; Microprocessors; Quality assessment; Sensors; Subsystems; Surveillance; Surveillance systems; Tracking systems; Video data
• ISSN: 2226-1494; 2500-0373
• DOI: 10.17586/2226-1494-2016-16-4-670-677

Subject of Research. The paper deals with the problem of multiple face tracking in a video stream. The primary application of the implemented tracking system is the automatic video surveillance. The particular operating conditions of surveillance cameras are taken into account in order to increase the efficiency of the system in comparison to existing general-purpose analogs. Method. The developed system is comprised of two subsystems: detector and tracker. The tracking subsystem does not depend on the detector, and thus various face detection methods can be used. Furthermore, only a small portion of frames is processed by the detector in this structure, substantially improving the operation rate. The tracking algorithm is based on BRIEF binary descriptors that are computed very efficiently on modern processor architectures. Main Results. The system is implemented in C++ and the experiments on the processing rate and quality evaluation are carried out. MOTA and MOTP metrics are used for tracking quality measurement. The experiments demonstrated the four-fold processing rate gain in comparison to the baseline implementation that processes every video frame with the detector. The tracking quality is on the adequate level when compared to the baseline. Practical Relevance. The developed system can be used with various face detectors (including slow ones) to create a fully functional high-speed multiple face tracking solution. The algorithm is easy to implement and optimize, so it may be applied not only in full-scale video surveillance systems, but also in embedded solutions integrated directly into cameras.