Crowd Counting Model Using Convolutional Neural Network

Abstract

Crowd Counting is being used for public safety, effective management of the crowd in elections or pilgrimages, music concerts. Recently there was a stampede in January 2022 in a temple in India where due to overcrowd many people were killed. Also, to stop the spread of pandemic crowd counting is proving to be beneficial in public places. Counting manually is a tedious task and it may produce false results, since it takes a long time. In crowded photos, objects appear to be partially surrounding each other as the density of people increase in the frame. Crowd counting is having limitations such as occlusion and background clutter. To solve this difficulty, earlier approaches relied on labelling complex density maps to understand the scale variation implicitly. Data preparation can be time expensive, and training these deep models might be problematic owing to a shortage of training data. As a result, we suggest an alternate and new method for crowd counting. The proposed model in this paper counts the number of people in the given image using a convolutional neural network based on ResNet50.