Crowd density analysis using subspace learning on local binary pattern

Crowd density analysis is a crucial component in visual surveillance for security monitoring. This paper proposes a novel approach for crowd density estimation. The main contribution
of this paper is two-fold: First, we propose to estimate crowd density at patch level, where the size of each patch varies in such way to compensate the effects of perspective
distortions; second, instead of using raw features to represent each patch sample, we propose to learn a discriminant subspace of the high-dimensional Local Binary Pattern
(LBP) raw feature vector where samples of different crowd density are optimally separated. The effectiveness of the proposed algorithm is evaluated on PETS dataset, and the results
show that effective dimensionality reduction (DR) techniques significantly enhance the classification accuracy. The performance of the proposed framework is also compared to other frequently used features in crowd density estimation. Our proposed algorithm outperforms the state-of-the-art methods with a significant margin.