A Privacy-Aware UAV Parking Framework with Polygon-Based Slot Association and Full-Frame Anonymization

Abstract

Large outdoor parking facilities are difficult to monitor with fixed sensors, whereas unmanned aerial vehicles (UAVs) provide wide-area coverage without per-space hardware. Converting aerial imagery into operational records, however, requires more than vehicle detection: detections must be assigned to specific parking spaces, linked to stored locations, and anonymized before retaining frames. We present a modular privacy-aware UAV parking framework combining YOLOv8s detection, box-normalized polygon association, geo-referenced slot records, and an independent full-frame anonymization branch. Because anonymization operates on the complete frame, sensitive regions remain eligible for processing regardless of occupancy-detector output. The four-stage occupancy detector achieved held-out test precision of 0.984, recall of 0.986, and mAP@0.5 of 0.963. At the predefined association threshold (𝜏 = 0.50), the mapping module correctly resolved 1,697 of 1,712 annotated slots, achieving 99.12% slot-state accuracy (95% image-cluster bootstrap CI: 98.34–99.66%). Under calibration-separated evaluation, end-to-end anonymization recall reached 91.37% for 6,015 WIDER FACE instances and 92.24% for 902 license plates, compared with 58.27% and 37.47% for same-data baselines. These results establish a traceable image-to-record pipeline in which occupancy detection, slot association, and anonymization are evaluated as distinct components.

Received: 3 June 2026 | Revised: 28 August 2026 | Accepted: 20 September 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in VisDrone-DET2019 [GitHub] at https://github.com/ VisDrone/VisDrone-Dataset, in HIT-UAV [GitHub] at https://github.com/suojiashun/HIT-UAV-Infrared-Thermal-Dataset, in the parking-specific occupancy dataset [GitHub] at https://github.com/Dineshmanchala19/parking-space-detector, in PKLot at https://web.inf.ufpr.br/vri/databases/parking-lot-database/, in WIDER FACE at https://shuoyang1213.me/WIDERFACE/, in the License Plate Object Detection benchmark [Hugging Face] at https://huggingface.co/datasets/keremberke/license-plate-object-detection, in yolov8-face [GitHub] at https://github.com/lindevs/yolov8-face, and in Hugging Face at https://huggingface.co/Babblu2821/alpr-plate-detector.

Author Contribution Statement

Aisha Alahmed: Software, Investigation, Data curation, Writing – original draft, Visualization. Saed Alqaraleh: Conceptualization, Writing – review & editing, Supervision. Qanout Alsharida: Methodology, Software, Validation, Data curation, Writing – original draft. Mariam Alahmed: Software, Formal analysis, Writing – original draft. Leen Boodai: Software, Validation, Writing – original draft. A. H. Abdul Hafiz: Conceptualization, Methodology, Writing – review & editing, Supervision, Project administration, Funding acquisition.