Deep Learning-Based Crowd Surveillance and Density Estimation

Crowd counting and surveillance are crucial for maintaining public safety and managing large gatherings effectively. Existing methodologies primarily include manual counting, traditional image processing, and various machine learning techniques. In this paper, we propose an enhanced deep learning-based methodology for more accurate crowd counting. Our study evaluated three human detection models: YOLO v3 with Canny Edge Detector, YOLO v3 with Hungarian & Kalman filtering and YOLO v5 with OpenCV functionalities. Among these, YOLO v3 with Canny Edge Detector demonstrated superior performance, achieving the highest accuracy (86.8%) and precision (75%). These results highlight the effectiveness of the YOLO model for crowd counting applications. Our findings indicate that advanced deep learning models significantly improve crowd surveillance and density estimation, offering promising implications for enhancing public safety measures and event management strategies.