Multi-Convolutional Neural Network Based Crowd Counting Using VGG-19 Architecture

Crowd counting has raised a major issue in computer vision with its extensive range of applications, like crowd management, public protection, and city development. The proposed model involves developing a method that can precisely determine the total number of crowds from a raw photograph captured from a random viewpoint and crowd density. A density map-based estimation that uses Multi-convolution Neural Network that is built on the VGG-19 architecture is developed that are trained to categorize the visual inputs with varying densities. The deeper architecture allows VGG-19 to learn more complex features from the input image, which is especially beneficial in crowd counting tasks where the input image contains a huge number of objects. The proposed crowd counting technique combines the VGG-19 architecture with the MCNN model to achieve accurate results. By using filters of different sizes, the MCNN model can handle input images of varying sizes or resolutions, while also accounting for variations in individuals or head sizes caused by viewpoint effects. This allows to accurately estimate crowd density without needing to identify the perspective map of the input image, making it highly versatile and operative for crowd counting tasks.