New approach to people density estimation based on video surveillance

摘要

This paper presents an algorithm for precise estimation of people density in indoor and outdoor areas. Automatic generation of so-called people density maps is based on video sequences acquired by surveillance systems. Our approach offers two types of estimation algorithms. The first one increments accumulation matrix when a person is detected in a given location. The second algorithm increments the second accumulation matrix only in the case of detecting a new person. Proposed methods were tested with the use of PETS 2009 database and our database of long-term video recordings. Finally, the results of visualization and determination of the “busy hours” are presented.