Locating Vehicles in a Parking Lot by Image Processing

We propose a method of detecting the existence of parked vehicles by processing the image of a parking lot taken by a surveillance camera. Whenever we want to park a car at a parking lot, how to find a proper parking division there causes a serious problem. The objective of the present article is in providing drivers with such information as the lot is fully occupied or relatively vacant, where unoccupied parking divisions are found, and so on. We employ images, since all areas in the parking lot can be observed with relatively few cameras, the system is compact, and the cost is not expensive. The image of a parking lot is taken by a surveillance camera set at some height in the parking lot. The relevant issues are how to cope with both tempo. ralandspatial changes in illumination, how to discriminate shadows from vehicles, how to cope with occlusion, and how to cope with various surface reflectances of vehicles, etc. To cope with these issues we transform the gray levels of the input images with log-transform, extracts edges and counts the number in each parking division, and then decides if each division is occupied or not. The recognition rate for a set of images taken at various moments of a day were well above 95 % .