Estimating the occupancy status of parking areas by counting cars and non-empty stalls

This work presents and compares different vision-based approaches to estimate the occupancy status of parking areas by counting cars and non-empty parking stalls. Our investigation considers both the scenario in which parking stalls are marked on the ground and the more challenging one in which no assumption on the presence or position of stalls is assumed. We carry out an experimental analysis on a real-world dataset of videos collected in different parking areas. Specifically, this work compares solutions based on image classification, vehicle detection and semantic segmentation. Our analysis highlights that: (1) methods based on image classification can be effectively leveraged when the position of parking stalls is known in advance, (2) methods based on image segmentation should be preferred over methods based on object detection when the geometry of the scene is not known, (3) temporal smoothing can be effectively used to improve predictions over time.