<p>Reliable traffic analysis, safety assessment, and urban mobility planning increasingly require multi-class and disaggregated traffic data, particularly in regions characterized by mixed traffic conditions and limited sensing infrastructure. Based on recent advances in unmanned aerial vehicles (UAVs) and computer vision that enable flexible, intersection-wise traffic data collection, this paper presents a UAV-based framework for the automated collection of multi-class traffic data, enabling class-specific counting of standard and adapted motor vehicles, and pedestrians, including cyclists. The framework integrates deep learning–based object detection with multi-object tracking to extract disaggregated traffic observations from aerial imagery. To support this study, a new dataset, YucaMex-Drone, was created, comprising 4,448 high-resolution UAV images annotated into eight traffic categories: cars, trucks, buses, vans, motorcycles, bicycles, moto-taxis, and pedestrians. These videos were collected at multiple urban and semi-rural intersections in Yucatán, Mexico. The performance of object detection models based on Faster R-CNN and YOLO architectures is evaluated, along with three counting strategies using different tracking approaches. Experimental results show that the best-performing detection model achieves a mean average precision of 98.15%, while the complete counting system attains a weighted precision of 82.79% across traffic classes. Beyond counting accuracy, the extracted data reveal spatial and temporal variations in multi-class traffic objects composition, highlighting the presence of vulnerable and informal road users often absent from conventional datasets. The proposed framework demonstrates how UAV-based visual sensing can be used as a traffic data acquisition tool to contribute to research on intelligent transportation systems.</p>