High-Resolution Human Density Monitoring via UAVs Using YOLOv8 and DeepSORT
Gabi Levin Dietrich, August Tabea Schwinghammer, Robert Waltraud Teufel, Zeynep Sahutoglu · 2025
This study presents a real-time human density analysis system based on unmanned aerial vehicles (UAVs), integrating deep learning techniques for object detection and tracking. The proposed platform combines the YOLOv8 algorithm for fast and accurate human detection with the DeepSORT algorithm for reliable identity-based tracking. A series of field tests were conducted under varying environmental conditions and crowd density levels to evaluate the system’s performance. High-resolution imagery collected by the drone was processed through a series of preprocessing steps to enhance detection quality, and spatial-temporal metadata was used for behavioral analyses and regional density estimation. The system achieved a precision rate of $93.6 \%$, a recall rate of $90.1 \%$, and a mean average precision ([email protected]:0.95) of 87.0%, demonstrating strong generalization capabilities in drone-based surveillance. In addition, tracking accuracy reached $88.6 \%$ with a low ID switch rate of approximately $5 \%$. The system operated at an average frame rate of 25 FPS with a latency of 120 milliseconds, confirming its suitability for real-time applications. Overall, the developed platform offers a robust, scalable, and practical decision-support tool for applications in public safety, urban planning, event management, and disaster response. Future work will explore advanced tracking models and edge computing architectures to further enhance system performance and autonomy.