Determining car-park occupancy from single images

We propose a system to estimate the occupancy of a car-park using a single image of a single camera. Very often car-parks are already equipped with CCTV-cameras for surveillance purposes which may be used for automatic detection systems as well. Our system is targeted on cases where occupancy values are sought, but exact solutions like automatic gates or induction loops are too costly and where estimate values are acceptable for the operator. The image processing for the vehicle classification basically works by constructing a reference image of the empty car-park given in the input image and then comparing those two. The occupancy estimate is determined by the vehicle to car-park pixel area ratio, where perspective distortion and occlusion is compensated.