Semantic Video Analytics Framework using SpinalXNet for Crowd Management in Large Public Events

In recent years, crowd management (CM) in large public events, such as concerts, sports gatherings, and festivals, has become increasingly critical for ensuring safety and preventing accidents. This task becomes more challenging owing to the high crowd density, severe occlusions, varying camera perspectives, and complex motion patterns that limit the reliability of traditional surveillance systems. However, conventional monitoring methods struggle to provide a real-time analysis and semantic understanding of crowd behavior. Hence, this study proposes a Semantic Video Analytics Framework (SVAF) for crowd management in large-scale public events. The framework integrates a SpinalNet-based Convolutional Neural Network (CNN) for feature extraction and density map generation to estimate crowd distribution, enabling an accurate assessment of crowd density and spatial patterns. Semantic modeling is applied to map low-level video features onto high-level contextual knowledge. The proposed SVAF-CM was evaluated using the UCF_CROWD_50 and UCF_QNRF datasets, where preprocessing included normalization, perspective correction, and noise reduction. Experimental results show that the SVAF-CM method achieves significant improvements in Mean Absolute Error (MAE) of 0.01 and Mean Squared Error (MSE) of 0.012 compared to the existing DCNN model.