UAV Remote Sensing and Deep Learning for Automatic Car Detection and Parking Occupancy Analysis: The Case of UANL Stadium, Mexico

Abstract. Rising traffic demand around university campuses and sports venues exacerbates parking scarcity and congestion. This study develops a UAV–deep learning workflow for the automatic quantification of parked vehicles and the estimation of occupancy levels across facilities at the Universidad Autónoma de Nuevo León (UANL). UAV surveys of the East and West Estadio UANL lots and the FIME faculty lots were conducted with DJI Mavic 2 and Matrice 350 RTK platforms during high-demand periods, including football matches and student egress peaks. The imagery, processed into centimeter-scale orthomosaics (2.4–2.8 cm ground sampling distance), enabled reliable instance detection using a pretrained Mask R-CNN Car Detection model. A total of 4,591 vehicles were identified across the surveyed areas: 2,336 in the West lot, 1,684 in the East lot, and 571 in the FIME lots. Kernel density estimation and spatial metrics revealed near-saturation of stadium lots during matches, reduced occupancy during off-event periods, and elevated but distributed demand in faculty lots during class dismissal. These geospatial indicators were integrated into a parking management framework using heat maps and bottleneck detection around access and egress roads. The approach demonstrates the potential of UAV–deep learning workflows to support demand-responsive parking control, traffic guidance, and long-term planning in congested university and event-driven environments.