CALCULATION AND ANALYSIS OF MOVING OBJECT FEATURES FOR TRACKING IN VIDEO SEQUENCES
Article Sidebar
Main Article Content
Abstract
This paper discusses the formation of object features considering the peculiarities of their presentation in video sequences. The main types of movement of a single object and object group are presented. We propose a classification of features that characterize the movement of objects in a video sequence. A modification of the algorithm for tracking multiple people on video sequences using the Kalman filter for outdoor video surveillance is described. The first stage requires detecting person in the input frames by YOLOv4 convolutional neural network. For assignment problem solving of person we store information about individual object in spatial domain of frames and in the time domain on a video sequence. For person description feature set is used: neural network and histogram features, center coordinates of a person in the frame, offset in the current frame relative to the previous one, person width and height in the previous frame, trajectory and time of movement. The results of experiments for video sequences obtained using a stationary and moving video camera are presented.
Article Details
This work is licensed under a Creative Commons Attribution 4.0 International License.
References
Клетте, Р. Компьютерное зрение. Теория и алгоритмы / Р. Клетте ; пер. с англ. А.А. Слинкин. – М. : ДМК Пресс, 2019. – 506 с.
Гонcалеc, Р. Цифровая обработка изображений / Р. Гонcалеc, Р. Вудс. –3-е изд. – М. : Техносфера, 2012. – 1104 с.
Sonka, M. Image Processing, Analysis, and Machine Vision / M. Sonka, V. Hlavac, R Boyle. – 4th ed. – Cengage Learning, 2015. – 930 p.
Application of Integral Optical Flow for Determining Crowd Movement from Video Images Obtained Using Video Surveillance Systems / H. Chen [et al.] // J. of Appl. Spectrosc. – 2018. – Vol. 85, iss. 1. – P. 126–133.
Histograms of oriented optical flow and Binet-Cauchy kernels on nonlinear dynamical systems for the recognition of human actions / R. Chaudhry [et al.] // IEEE Conference on Computer Vision and Pattern Recognition (CVPR). – 2009. – 1932–1939.
Motion Maps and Their Applications for Dynamic Object Monitoring / H. Chen [et al.] // Pattern Recogn. Image Anal. – 2019. – 29. – P. 131–143.
Богуш, Р.П. Алгоритм сопровождения людей на видеопоследовательностях с использованием сверточных нейронных сетей для видеонаблюдения внутри помещений / Р.П. Богуш, И.Ю. Захарова // Компьютерная оптика. – 2020. – Т. 44, № 1. – С.109–116.
Bochkovskiy, A. YOLOv4: Optimal Speed and Accuracy of Object Detection [Electronic resource] / A. Bochkovskiy, Ch.-Y. Wang, H.-Y. M. Liao. – Mode of access: https://arxiv.org/abs/2004.10934. – Date of access: 12.08.2020.
Simon, D. Optimal State Estimation: Kalman, H Infinity, and Nonlinear Approaches / D. Simon. – New Jersey : John Wiley & Sons, 2006.
MOTChallenge: The Multiple Object Tracking Benchmark [Electronic resource]. – Mode of access: https://motchallenge.net. – Date of access: 16.06.2020.
Person Re-ID (PRID) Dataset [Electronic resource]. – Mode of access: https://www.tugraz.at/institute/icg/research/team-bischof/lrs/downloads/prid11/. – Date of access: 12.04.2019.
iLIDS Video re-IDentification (iLIDS-VID) Dataset [Electronic resource]. – Mode of access: http://www.eecs.qmul.ac.uk. – Date of access: 12.04.2019.
Wojke, N. Simple online and real time tracking with a deep association metric / N. Wojke, A. Bewley, D. Paulus // IEEE International Conference on Image Processing 2017: 3645-3649. – DOI: 10.1109/ICIP.2017.8296962.
Real-time Multi-person tracker using YOLOv3 and deep_sort with tensorflow [Electronic resource]. – Mode of access: https://github.com/Qidian213/deep_sort_yolov3. – Date of access: 16.06.2019.