Crowd density estimation with a block-based density map generation

Crowd management is one of the challenging tasks in computer vision especially crowd counting which can be the key solution for many surveillance applications. But the estimation of crowdedness in a scene can be related to many problems that limit the effectiveness of any method, we can cote from the theme the scale variation of the objects, and the similarity between the background and the foreground in some complex scenes, as well as the variation of the degree of crowdecity within the same analyzed data. In this paper, we propose a block-based crowd counting model by collaborating the VGG layer with channel-wise attention modules between each block of layers (Crowd-per-Block). the channel attention is used to distinguish between the background and foreground texture. At the end of the network and to extract the contextual information and capture the change in density distribution we introduced a cascaded-spatial-wise attention module. The proposed method is evaluated on various datasets. The experimental results show that the proposed method works well for fully crowded scenes while it's less accurate for less crowded scenes.