Abstract
Visual understanding of crowd scenes is a challenging and important issue in computer vision domain. Identification of crowd type is a basic requirement for analyzing crowd scenarios. With the advancement of deep convolution neural networks image recognition problems have become easy. In this paper, we propose a novel architecture (DeepCrowd) inspired by Resnet to incorporate spatial features comprehensively. To train and evaluate proposed system, a robust and unique dataset of nearly six thousand images is generated. Evaluating the system extensively highlighted accuracy of 83.11% that is comparable with others state-of-the-art methods.