Information Technology Reference
In-Depth Information
7 Conclusion and Future Work
In this paper, different color descriptors, an innovative technique for smoothing
color difference between different FOVs and unsupervised classifiers are studied
in the football multi-views environment. We evaluated three different color de-
scriptors (RGB, normalized RGB and Transformed RGB histograms) we trans-
formed them by a CBTF in order to mitigate the FOVs difference, and three
unsupervised classifiers (MBSAS, BCLS and k-means). Other descriptors, as
Color Sift and Moments, have not been considered since they are not reliable in
presence of highly deformable objects, such as moving players. After the experi-
ments on real videos, we can conclude that the better performance were carried
out using the Transformed RGB histograms combined with k-means classifier af-
ter the application of CBTF on the original data. As a future work, the analysis
of unsupervised team discrimination here proposed could be further improved by
considering different feature sets, and different classifiers. One weak point of our
experiments was that similar uniforms can be seldom found and all the methods
suffer in separating different classes. This results was expected since the consid-
ered color descriptors are based on histogram evaluations that lose the spatial
information on the color distribution. The next step will be to investigate on
color features that can be applied to highly dynamic moving objects, not sub-
ject to rigid motion constraints, such as connected graphs of color histograms or
weighted histograms on segmented body parts.
References
1. Assfalg, J., Bestini, M., Colombo, C., Del Bimbo, A., Nunziati, W.: Semantic an-
notation of soccer videos: automatic highlights identification. CVIU 92, 285-305
(2003)
2. Beetz, M., Bandouch, J., Gedikli, S.: Camera-based observation of football games
for analyzing multi-agent activities. In: AAMAS, pp. 42-49 (2006)
3. Ekin, A., Tekalp, A., Mehrotra, R.: Automatic soccer video analysis and summa-
rization. IEEE Trans. on Image Processing 12, 796-807 (2003)
4.Hayet,J.,Mathes,T.,Czyz,J.,Piater,J.,Verly,J.,Macq,B.:Amodulamulti-
camera framework for team sports tracking. In: AVSS, pp. 493-498 (2005)
5. Mazzeo, P.L., Spagnolo, P., d'Orazio, T.: Object tracking by non-overlapping dis-
tributed camera network. In: Blanc-Talon, J., Philips, W., Popescu, D., Scheunders,
P. (eds.) ACIVS 2009. LNCS, vol. 5807, pp. 516-527. Springer, Heidelberg (2009)
6. Naemura, N., Fukuda, A., Mizutani, Y., Izumi, Y., Tanaka, Y., Enami, K.: Morpho-
logical segmentation of sport scenes using color information. IEEE Tr. on Br. 46,
181-188 (2003)
7. Prosser, B., Gong, S., Xiang, T.: Multi camera matching using bi-directional cu-
mulative brightness transfer functions. In: BMVC 2008 (2008)
8. Spagnolo, P., D'Orazio, T., Leo, M., Distante, A.: Moving object segmentation
by background subtractionand temporal analysis. Im. Vis. Comp. 24(5), 411-423
(2006)
9. Theodoridis, S., Koutroumbas, K.: Pattern Recognition. Academic Press, London
ISBN 0-12-686140-4
Search WWH ::
Custom Search