GeoPython 2024: GeoAI in Action Leveraging Python for Advanced Parking Spot Detection

Nils Hamel
GeoAI in Action: Leveraging Python for Advanced Parking Spot Detection

Using Python and GeoAI, we developed an AI model to efficiently map out parking spots from aerial images of a large French metropolitan area, significantly simplifying the city's parking inventory management.

The project commenced with the objective of detecting parking spots and parking lots within a large French metropolitan (60 square km area). The metropolitan authorities are mandated to establish and maintain an inventory of available parking spots in the public spaces of the city streets. This task is time-consuming and demands significant human resources and effort. To address this, our proposal suggested utilizing aerial images and AI solutions to create an inventory of parking spots. This approach enables the development of a solution that can be applied to future aerial images, simplifying the maintenance of the inventory.

We trained a deep learning model on aerial images to detect parking spots, employing instance segmentation and a python framework using pytorch and Mask-RCNN networks. Python tools were also heavily used to prepare data and to analyze geospatial results using geopandas, rasterio and other Python modules. The diversity in parking spot appearances posed a challenge during training, but ultimately yielded satisfactory results. Moreover, we implemented a Python-based filtering processes post-AI solution to enhance the quality of the inventory obtained. Upon project completion, we delivered a solution comprising AI tools and a traditional process capable of extracting a parking spot inventory from large aerially captured areas. Subsequently, we estimated the quantity of individual parking spots by analyzing parking spot properties through inference.