A Computationally Efficient Tracking Scheme for Localization of Soccer Players in an Aerial Video Sequence

Abstract

The authors attempted to construct a novel sensor networking system that estimates locations of sensor nodes as locations of humans wearing them via image processing. In this application, computationally efficient human localization is indispensable because the operation should be executed on embedded systems. To actualize human localization at a lower computational cost, this study proposes a novel tracking scheme using only detection results. Experimental results using the computer graphics (CG) dataset that was created using player locations of an actual soccer game showed that the proposed scheme outperformed existing schemes implemented in the OpenCV library.