A NEURAL NETWORK APPROACH FOR COUNTING NUMBER OF IMAGES

Crowd counting is a dominant topic of research in the field of computer surveillance, but if the crowd distribution is unequal the perfection of crowd counting on the basis of the MCNN has yet to be improved to adapt to uneven crowd distributions MCNN has been used to perform aggressive results. The global density feature for crowds in a project is taken into account, and globalised density aspects are drawn and added to the MCNN by the cascaded learning method, because in the course of the down-sampling process certain comprehensive features will be disregarded by the MCNN, and this will affect the precision of the density map. The experimental results for ucfcc50 and the shanghaitech data set indicate that the method proposed is more accurate and more stable