MRCNet: Crowd Counting and Density Map Estimation in Aerial and Ground Imagery

In spite of the many advantages of aerial imagery for crowd monitoring and at mass events, datasets of aerial images of crowds are still in the field. As a remedy, in this work we introduce a novel crowd, the DLR Aerial Crowd Dataset (DLR-ACD), which is composed of 33 large images acquired from 16 flight campaigns over mass events with 226,291 annotated. To the best of our knowledge, DLR-ACD is the first aerial dataset and will be released publicly. To tackle the problem of accurate counting and density map estimation in aerial images of crowds, this work proposes a new encoder-decoder convolutional neural network, the so-called-Resolution Crowd Network MRCNet. The encoder is based on the VGG-16 and the decoder is composed of a set of bilinear upsampling and layers. Using two losses, one at an earlier level and another at last level of the decoder, MRCNet estimates crowd counts and-resolution crowd density maps as two different but interrelated tasks. In, MRCNet utilizes contextual and detailed local information by high- and low-level features through a number of lateral connections by the Feature Pyramid Network (FPN) technique. We evaluated MRCNet on proposed DLR-ACD dataset as well as on the ShanghaiTech dataset, a-based crowd counting benchmark. The results demonstrate that MRCNet the state-of-the-art crowd counting methods in estimating the crowd and density maps for both aerial and CCTV-based images.