Abstract—This study proposes a method for automated detection of parking space occupancy in environments with absent or unclear markings. The system uses a U-Net11-based image segmentation model supported by a preprocessing pipeline that includes contrast correction, perspective transformation, and Contrast Limited Adaptive Histogram Equalization (CLAHE) enhancement to improve feature visibility under varying lighting conditions. The model was trained on a combined dataset from surveillance cameras and open-source images. On the test set, it achieved an accuracy of 74%, with Precision of 75.4%, Recall of 70.1%, F1-score of 72.6%, and Intersection over Union (IoU) of 57%. The Precision-Recall operating point (Precision = 75.42%, Recall = 70.27%) confirms reliable classification in challenging scenes. Despite the moderate segmentation accuracy, the achieved 74% is sufficient for practical parking-space detection because the final occupancy decision is based on aggregated spatial predictions rather than pixel-level precision. Local segmentation errors do not significantly affect the binary classification of “occupied/vacant”, making the system robust to noise, weather variability, and imperfect markings. Experiments confirm that occupancy determination remains stable even when segmentation boundaries are slightly distorted. The proposed approach improves the efficiency of identifying available parking spaces and can be integrated into smart city infrastructure to support automated parking monitoring and urban mobility optimization.
 
Keywords—parking automation, parking occupancy map, computer vision, modified U-Net11 architecture, neural network model


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).