Reliability of bench-mark datasets for crowd analytic surveillance

This paper presents an evaluation on the reliability of the bench-mark datasets for outdoor crowd analytic surveillance systems. The credibility of the databases are assessed based on their diverseness to yield challenges of dynamic environments. The main object of this paper is to assess the challenges imposed by the databases for sudden illumination variance and effect of wavering trees. Two bench-mark databases, PETS 2010 and OTCBVS, along with our proposed dataset are evaluated using the three most popular background modelling algorithms in crowd analytic surveillance; Approximate Median Method, Gaussian Mixture Model and Codebook. The diverseness of these databases are assessed, with respect to the performance of the basic algorithms using qualitatively and quantitatively. Eventually the reliability of these bench-mark databases for outdoor crowed analytic surveillance is assessed.