Sports Type Classification Using Signature Heatmaps

Rikke Gade, Thomas B. Moeslund; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2013, pp. 999-1004

Abstract


Automatic classification of activities in a sports arena is important in order to analyse and optimise the use of the arenas. In this work we classify five sports types based only on occupancy heatmaps produced from position data. Due to privacy issues we use thermal imaging for detecting people and then calculate their positions on the court using homography. Heatmaps are produced by summarising Gaussian distributions respresenting people over 10-minute periods. Before classification the heatmaps are projected to a low-dimensional discriminative space using the principle of Fisherfaces. Our result using two weeks of video are very promising with a correct classification of 90.76 %.

Related Material


[pdf]
[bibtex]
@InProceedings{Gade_2013_CVPR_Workshops,
author = {Gade, Rikke and Moeslund, Thomas B.},
title = {Sports Type Classification Using Signature Heatmaps},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2013}
}