Dilated CNN for Crowd Count

In the current scenario of the COVID-19 pandemic estimating the count of number of people present in public places at a particular time has become a significant task. Crowd count is attracting a lot of researchers from the computer vision and deep learning field. It has been found that to achieve this objective computer vision techniques such as deep learning, machine learning, etc. outperform traditional ways of estimating crowd count that uses handcrafted features such as Histogram of gradients, Haar, Scale Invariant Feature Transform and gives better results with higher accuracy. The paper studies the effect of dilation on convolution layers in estimating the crowd count. We have also done a comparative analysis of the developed model with different dilation rates on the ShanghaiTech dataset (part A and part B). The model is trained with images containing occluded and restricted visibility of heads. The model outputs the result with substantial accuracy in estimating the headcount in images of the dense crowd in a sensibly less amount of time.