Unsupervised detection of available parking spots

In recent years, finding an available parking spot is becoming increasingly difficult, especially near crowded places. This leads to frustration and waste of time and fuel among many drivers. In this paper, we propose an efficient computer-vision algorithm for estimating parking availability, based on unsupervised learning of parking spot locations. The learning stage is based on analyzing the motion field due to car motion in parking areas, for estimating parking spot locations. A parametric curve is then fitted to the estimated locations, and parking spot centers are detected using the K-means algorithm. The status of each parking spot (i.e., vacant/occupied) is based on the distance of a detected entering/leaving car to the spot center. The algorithm was tested in several real-world scenarios, with no prior information on the parking spot location structure, exhibiting good detection performance.