Research on dense crowd counting method combining attention and hybrid dilated convolution

Abstract

To solve the problems of background interference, noise and occlusion in crowd counting in complex scenes, a CA-HDC crowd counting method was proposed. The network consists of the front end and the back end. The front end selects the first 10 layers of the VGG-16 network. In the back end, the SE attention mechanism is fused with the hybrid cavity convolution layer that alternately uses the expansion rate. The head features are weighted by the attention mechanism, and the receptive field of the network is expanded by the hybrid cavity convolution to eliminate the influence of background interference and noise. Experiments are carried out on two public data sets. Compared with other crowd counting networks, the proposed network achieves better results.