Crowd and Traffic Analysis in Drone Imagery Using AI/ML
Pages:7-73
Samruddhi Jagtap, Atharv Kulkarni,Janvi Adam, Purva Shingare, Vinay Nagalkar
Urbanization has significantly increased the complexity of crowd management and traffic control, necessitating the development of intelligent, automated surveillance systems. This paper proposes a comprehensive drone-based monitoring framework that leverages YOLOv8, a state-of-the-art deep learning algorithm, for that accurate detection and classification of pedestrians and vehicles. The model was trained on a curated dataset of 16,000 high-resolution aerial images collected from multiple urban scenarios, including busy intersections, public gatherings, and congested roadways. Data augmentation techniques such as rotation, flipping, scaling, and exposure adjustment were applied to enhance generalization and improve performance under varying lighting and perspective conditions. For deployment, the trained YOLOv8 model was ported to a Raspberry Pi 5 with 8 GB RAM and a connected Pi Camera v3 mounted on a quadcopter, enabling edge-based real- time inference at 18 frames per second without relying on cloud processing. The system achieved a mean average precision (mAP) of 85.3%, with class-wise precision of 92.8% for pedestrians and 95.2% for vehicles. An experimental evaluation across three urban locations, under diverse weather and lighting conditions, confirmed the robustness, low latency, and reliable operation of the system, even in high-density scenes. By combining on-edge AI processing with autonomous aerial mobility, the proposed framework improves response time, enhances data privacy, and offers scalable solutions for smart city applications. The study demonstrates the feasibility of deploying lightweight, high- accuracy object detection models on embedded platforms, marking a step toward autonomous, real-time aerial surveillance for urban safety management and intelligent traffic monitoring. Urbanization has increased the need for intelligent traffic and crowd monitoring systems. This study proposes a drone-based framework using YOLOv8 for accurate real-time detection of pedestrians and vehicles, deployed on a Raspberry Pi for edge processing. The system achieves high accuracy and low latency, making it suitable for scalable smart city surveillance applications.
Crowd Analysis, Traffic Monitoring, YOLOv8, Raspberry Pi 5, Drone Surveillance, Edge AI