Traffic flow modeling is essential for transportation engineering and urban planning, particularly in complex scenarios such as lane-changing and movements at ramps. However, obtaining high-quality trajectory data remains challenging, especially in urban environments where traditional methods like stationary cameras fall short. Existing drone-based datasets often lack full coverage of traffic states and critical merging and diverging behaviors at ramps. This study presents MiTra (Milan Trajectories), a high-resolution traffic trajectory dataset collected using six drones over a 900 m section of the A50 urban freeway in Milan, Italy. Spanning all traffic states from free flow to congestion, it captures detailed vehicle behavior at on-ramps and off-ramps through nine flight campaigns for 135 min. It includes 124 641 vehicle trajectories, averaging 650 m in length after stitching, with detailed positions, speeds, and accelerations. Nearly half of the vehicles executed lane changes. The dataset provides stitched trajectory data, raw drone videos, and tracking logs. Comprehensive quality checks, including vehicle detection and video stitching validation, ensure its reliability for traffic modeling, autonomous driving research, and computer vision applications.