Improving Crowd Counting Performance: A Convolutional Neural Network Approach with Transfer Learning

Precise crowd counting is critical to public safety and smart city planning since it solves the problems associated with the time-consuming manual counting of people in photos and videos. Transfer learning has become a key building block for improving crowd counting techniques, especially when used to Convolutional Neural Networks (CNNs). Because pretrained models already know the pertinent weights and architecture, using them in transfer learning minimizes computational demands and shortens training time. This paper presents a crowd counting method with an emphasis on optimizing the VGG16 model with a mall dataset. The results show that using VGG16 for transfer learning leads to higher performance when compared to more modern methods like AdaCrowd and PSSW models. In addition, the paper highlights how adaptable our proposed method is and how well it can transfer knowledge from one dataset to another.