Violent Behaviour Analysis in Crowd

In response to the increasing demand in crowd management, and to solve the challenges of categorizing crowds as violent and non-violent, we have introduced a new architecture called Violent Behaviour Analysis (VBA). The VBA architecture combines the Congested Scene Recognition Network (CSRNet) and the Unet, CSRNet consists of a data acquisition module in the foreground and a processing unit in the background. The data acquisition module captures intricate crowd-related details and the processing unit performs the generation of a density map for precise spatial representation. The Unet along with CSRNet estimates density maps and also distinguishes crowd regions from their backgrounds to improve accuracy. The architecture is evaluated using a video dataset known as the Violent dataset built from an image dataset of fine-grain crowd-counting. The results with density maps and CMAE demonstrate the efficiency of this approach in categorizing the crowd as violent and violent individuals.