Performance assessment of automatic crowd detection techniques on airborne images

Abstract

Real-time monitoring of crowded regions has crucial importance to avoid overload of people in certain areas. Understanding dynamics of large people crowds can also help to estimate future status of public areas. In order to bring an automatic solution to the problem, herein we introduce four different approaches based on feature extraction from airborne images. Using four different feature extraction methods separately, we form four different probability density functions (pdf) which holds information about density of people. With our experimental results, we discuss the strengths and weaknesses of these methods in detail. Besides usage of four different feature extraction methods, we also introduce feature fusion approaches in order to increase robustness of the algorithm. Our promising experimental results indicate possible usage of the algorithm on real-time on board applications.