Abstract
Crowd counting is an important task for crowd monitoring in Masjid Al-Haram, where millions of people gather every year to fulfil religious obligation. Several strides have been made to automatically estimate the density and count from images. However, it still remains a challenging task due to variations in view points, scales and illumination. In this paper, we propose a novel approach for the crowd counting based on Convolutional Neural Network (CNN). In this approach, we first divide the input image into non-overlapping blocks and then each block is further sub-divided into cells. For each cell, we extract corresponding patch in the image and then feed to CNN. We then train a binary CNN classifier, which classifies each patch into two classes, i.e, head or background. We evaluate our method on our own dataset which we collected from different location of Masjid Al-Haram. From the experiments, we show to achieve 90% accuracy. We compare our proposed method with other state-of-the-art methods and from the experimental results, we show that our proposed method outperforms other state-of-the-art methods