As urban populations continue to expand and vehicle density increases, the need for effective traffic monitoring and management becomes increasingly critical. Urban areas face significant challenges such as growing congestion, safety concerns, and parking shortages, all of which contribute to reduced mobility, elevated pollution, and prolonged travel times.

Conventional traffic monitoring systems, often reliant on fixed road cameras, suffer from limited coverage, privacy issues, and operational inflexibility. Unmanned Aerial Vehicles (UAVs), by contrast, offer advantages such as wider coverage, greater operational flexibility, and cost-effectiveness, making them well-suited for macroscopic monitoring of urban road networks. Concurrently, the surge in traffic data availability underscores the potential of machine learning in traffic analysis. This dissertation presents novel approaches that integrate UAV technology and machine learning to address three critical aspects of urban traffic management: traffic density estimation, lane-change prediction, and parking lot monitoring.

First, this work introduces the pNEUMA Vision dataset, a high-resolution UAV-captured visual dataset designed to advance traffic analysis. The dataset is a valuable resource for developing data-driven models and serves as a foundation for tackling a wide range of urban traffic challenges explored in subsequent chapters.

Second, the dissertation explores machine learning applications for traffic data analysis, leveraging the pNEUMA Vision dataset, with a focus on traffic density estimation and lane-change prediction. A novel vehicle counting method, based on density maps, is proposed for urban arterial roads. Unlike traditional vehicle detection methods, this approach leverages density maps to estimate vehicle numbers from aerial imagery, offering superior performance in identifying congestion patterns and vehicle queue formations. Additionally, a lane-change prediction model utilizing a transformer-based architecture is developed to forecast lane-change events in dense urban traffic. By incorporating surrounding vehicle dynamics and addressing the rarity of lane changes with class balance focal loss and two-stage training, the model achieves notable accuracy improvements. Performance is further enhanced through the inclusion of the Dynamic Time Warping (DTW) metric for refined assessment. These innovations provide deeper insights into urban traffic dynamics, ultimately contributing to more efficient urban traffic flow management.

Lastly, the dissertation proposes an automated UAV-based system for monitoring both on-street and off-street open parking. The system reduces the need for manual data annotation through a novel pseudo-labeling method, which is enhanced by variations in UAV viewpoints. Experimental results from a city in Switzerland demonstrate the system's effectiveness, including its ability to monitor parking turnover rates, offering a practical solution for improved urban parking management.

In summary, this dissertation contributes to the advancement of Intelligent Transportation Systems (ITS) by delivering scalable, data-driven solutions that integrate UAV technology with machine learning. These contributions address key urban traffic challenges, including real-time traffic monitoring and vehicle behavior prediction, with the potential to significantly enhance the sustainability and efficiency of urban transportation systems.