Classification algorithm of parking space images based on a histogram of oriented gradients and support vector machines

In this paper, a classification algorithm of parking space images is proposed to improve the accuracy of parking space classification, which can be used in smart parking management systems based on video surveillance. The descriptors of a parking space image are formed on the basis of a histogram of oriented gradients by performing the following steps: computation of vertical and horizontal gradients of the original parking space image, computation of the modulus of the gradient and orientation vectors, the gradients are then accumulated into separate cells according to their orientation, the cells are united into blocks, and the orientations of block's cells are normalized. A support vector machine is used to classify the descriptors of the parking space. The purpose of the research was to determine the most efficient parameters of the parking space descriptor and a kernel function. The paper presents the results of experiments.