Current - Issue

This work presents a vision-based approach for people counting and proximity- based risk analysis using deep learning and computer vision techniques. The proposed system utilizes YOLO (You Only Look Once) to accurately detect and localize individuals in an image by generating bounding boxes. The number of detected persons is computed to estimate crowd size, while centroid-based distance calculation is employed to analyze spatial relationships between individuals. A threshold-based mechanism identifies proximity violations, enabling the system to detect unsafe interactions within the scene.In addition to detection and distance analysis, the system evaluates key metrics such as the number of violations, the number of people at risk, and the violation percentage to classify the overall risk level into categories such as low, medium, and high. A density heatmap is also generated to provide a visual representation of crowd distribution. The system is implemented with a user-friendly graphical interface that enables easy image input and real-time analysis. Experimental results demonstrate that the proposed approach effectively performs people detection, counting, and risk assessment across different crowd scenarios. However, limitations such as pixel-based distance measurement and sensitivity to perspective distortion are acknowledged. The proposed method offers a practical and efficient solution for applications in surveillance, public safety, and smart city environments.