Object Tracking in the Video Stream by Means of a Convolutional Neural Network

Yu. N. Zolotukhin 1 K. Yu. Kotov 2 * A. A. Nesterov 1 E. D. Semenyuk 1

1Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences, Novosibirsk, 630090 Russia

2Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences,Novosibirsk, 630090 Russia

Correspondence to: *e-mail: kotov@idisys.iae.nsk.su

March 3, 2020

Abstract—A new algorithm of 6-coordinate tracking of a moving object on a sequence of RGB-images that is based on the convolutional neural network is proposed. Training the neural network is carried out by using the synthesized data of the object with a dynamic model of motion. A Kalman filter is included into the feedback from the network output to its input to obtain a smoothed estimate of the object coordinates. Preliminary results of object tracking on synthesized images demonstrates the efficiency of the proposed approach.

Keywords: trackingvideo streamconvolutional neural networkKalman filter

DOI: 10.3103/S8756699020060163