People counting using combined feature

Abstract

In this paper, we present a new people counting approach in visual surveillance scenes. The features adopted in previous methods are all extracted at pixel-level or based on local area, which are severely affected by factors such as occlusion. To cover the shortage, we introduce a new feature which describes a people crowd as a whole. Because pedestrian behaviors change when the degree of crowdedness varies, we can capture motion information to model a crowd and characterize the pedestrian behaviors based on statistic analysis. Afterwards we combine together the two kinds of features presented above as the final people counting feature. Experiments conducted in real world scenes demonstrate the superior effectiveness of the proposed method.