Parking Space Detection in the City of Granada

TL;DR

The paper tackles parking-space detection in Granada by performing semantic segmentation of aerial imagery into parked cars, moving cars, and roads, and introduces a Granada-specific dataset (GranadaAerial) of 90 labeled images for model adaptation. It systematically compares FCN-based architectures—Dynamic U‑Net, PSPNet, and DeepLabV3+ with ResNet101 backbones—and finds DeepLabV3+ to be the strongest performer. To handle Granada-domain specifics, it develops two parked-car strategies: a heuristic post-processing approach (UnParked Car Model) and a learning-based four-class model (Parked Car Model), trained in two stages on UDD5/UAVid and then GranadaAerial. Evaluation uses Foreground accuracy, Dice Coefficient, and Jaccard Index, revealing DeepLabV3+ as the best baseline, while domain shift and data scarcity motivate dataset expansion and daily aerial analytics for urban parking planning and traffic management.

Abstract

This paper addresses the challenge of parking space detection in urban areas, focusing on the city of Granada. Utilizing aerial imagery, we develop and apply semantic segmentation techniques to accurately identify parked cars, moving cars and roads. A significant aspect of our research is the creation of a proprietary dataset specific to Granada, which is instrumental in training our neural network model. We employ Fully Convolutional Networks, Pyramid Networks and Dilated Convolutions, demonstrating their effectiveness in urban semantic segmentation. Our approach involves comparative analysis and optimization of various models, including Dynamic U-Net, PSPNet and DeepLabV3+, tailored for the segmentation of aerial images. The study includes a thorough experimentation phase, using datasets such as UDD5 and UAVid, alongside our custom Granada dataset. We evaluate our models using metrics like Foreground Accuracy, Dice Coefficient and Jaccard Index. Our results indicate that DeepLabV3+ offers the most promising performance. We conclude with future directions, emphasizing the need for a dedicated neural network for parked car detection and the potential for application in other urban environments. This work contributes to the fields of urban planning and traffic management, providing insights into efficient utilization of parking spaces through advanced image processing techniques.

Figures (5)

• Figure 1: Training and validation loss comparison across epochs for Dynamic U-Net, PSPNet and DeepLabV3+ models. The graph illustrates the trend of validation loss over 50 epochs, highlighting the stability of Dynamic U-Net and PSPNet and the occasional spikes in validation loss for DeepLabV3+, which recovers and maintains a leading performance in subsequent epochs.
• Figure 2: Comparison of Dice Coefficients for Image Segmentation during the first stage of the training.
• Figure 3: The figure displays the training and validation loss for both the Parked Car Model and the UnParked Car Model.
• Figure 4: Analysis of Parked Car Detection Methodologies in Aerial Imagery. The analysis showcases three distinct approaches across rows: the baseline detection of cars and roads (Row 1), the implementation of a heuristic algorithm for parked car identification (Row 2) and a model specifically trained to detect parked cars (Row 3). For each method, the columns display the original image, the ground truth segmentation, the algorithm's predicted segmentation and the error mask, respectively. The error mask uses red to signify missed detections (false negatives), green for incorrect detections (false positives) and white for correct detections (true positives).
• Figure 5: Performance evaluation of our model on images from Granada, exhibiting challenges in identifying cars and roads (left) and the corresponding predicted masks (right). Background is represented with black color, cars in blue color and roads in lilac color. The model's difficulty in detecting roads under shadows and identifying cars and roads at hte same scale suggests a domain shift and limited dataset diversity in the training phase.