Enhanced Crowd Counting Using Patch-Level Annotation with YOLO Deep Learning
Accurate crowd counting is of great importance to a wide range of real-world applications such as public safety monitoring, traffic management, event organisation, urban planning and emergency response. However, existing crowd counting techniques often suffer from severe occlusions, large variations in crowd density, perspective distortion and complex background environments, leading to reduced detection accuracy. In this paper, a new crowd counting framework is proposed based on patch-level annotation and YOLO (You Only Look Once) deep learning architecture to improve detection performance and real-time processing capability. The suggested method separates high-resolution crowd photos into smaller patches. It performs accurate patch-level annotation, allowing the model to learn local crowd characteristics, minimise annotation complexity, and better localise tightly packed people. Processing these image patches with YOLO's quick object detection allows reliable crowd estimation with low inference time. The proposed framework surpasses state-of-the-art approaches in accuracy, robustness, and computational efficiency, as demonstrated by extensive trials on benchmark crowd-counting datasets. A performance study using standard measures demonstrates that identification and counting accuracy improve significantly across diverse crowd densities and challenging conditions. The suggested system is effective and scalable for intelligent surveillance, smart city monitoring, public event management, and real-time crowd analysis.