Automated Crowd Detection in Stadium Arenas


In this paper, we present an approach for crowd detection based on an ensemble of classifiers which employ several feature representation schemes of crowd images, including, local ternary patterns, local binary patterns, and features based on the spatial gray level dependency matrix. A Support Vector Machine classifier is trained on each of these feature vectors. Classifier predictions are then combined by sum rule. Experiments are performed on a large dataset that contains challenging sequences recorded during real football matches at a stadium arena. Experimental results confirm that the different feature representations give complementary information which is exploited by fusion rules. The method proposed in this paper is shown to outperform previous methods tested on the same dataset. MATLAB code using the different descriptors is available at

Keywords crowd detection, local ternary patterns, spatial gray level dependency matrix, local binary patterns.

[full paper]