PARKING OCCUPANCY ESTIMATION ON SENTINEL-1 IMAGES
Keywords: Parking occupancy estimation, SAR, Sentinel-1, time-series, economic activity, COVID-19
Abstract. This paper presents a method to estimate the occupancy ratio of parkings from SAR satellite images. The algorithm takes as input a series of Sentinel-1 images along with a mask indicating where the parking is located and returns for each image an occupancy ratio. The method is generic and can easily be extended. We validate our results in two parts. First, we have created a dataset of Sentinel-1 GRD image time series where each image is associated to a ground truth parking occupancy ratio. This ground truth is estimated thanks to a surveillance camera that permanently films and records the parking. We observe a strong correlation between the estimated occupancy rate and the ground truth occupancy rate. Secondly, we estimate the occupancy ratio of the 250 largest retail parkings in France from January 2018 to April 2020. We observe that weekly and seasonal patterns are consistent with consumer and economic trends. Parking occupancy estimations also plummet during the COVID-19 containment measures.