Parking Space Detection Under Different Lighting Conditions: A Comparative Study of Yolo-Based Models

Accurate detection of parking spaces is essential for optimizing urban mobility and improving the efficient use of parking areas. Convolutional neural networks (CNNs) have become widely used tools for automating this task through image analysis. However, many architectures exhibit significant variations in performance under different lighting and weather conditions. In this work, we evaluate several deep learning--based object detection architectures for parking space detection under varying illumination scenarios. The experiments were conducted using a dataset containing images captured under different weather and lighting conditions. The performance of multiple models, including YOLO-based architectures and RetinaNet, was analyzed using the mean Average Precision (mAP) metric. The results show that YOLOv9 achieves the most consistent and superior performance across most scenarios, reaching a mAP value of 0.995. These results indicate that YOLOv9 represents a robust solution for detecting parking spaces in urban environments with variable lighting conditions.