Visión artificial para control de estacionamiento vehicular

Abstract

Currently one of the causes for the increase in traffic congestion is the lack of availability of vehicle parking. This research focuses on the development of a system capable of identifying available parking spaces by analyzing parking images and videos using the Open CV computer vision library and the Python programming language. It consists of identifying and defining the coordinates of the parking space region of interest (ROI). Using artificial vision, the detection of free or occupied space is carried out using the Gaussian blur technique, then converting the images from RGB to grayscale. Subsequently, the original ROI and the ROI of the converted image are evaluated and finally, calculate if the standard deviation and the average are greater and less than a threshold. The last activity of this research work is based on the development of five tests in different settings. The confusion matrix was applied to each test as a test evaluation tool and the metrics: Accuracy (Accuracy), Sensitivity (Recall) and Precision (Precision). The evidence obtained indicates that the proposed model allows detecting the state of the parking spaces within a parking lot, with a degree of accuracy of 92%, precision of 80% and a sensitivity of 79% as an average of the five tests carried out. These results also showed that the success of the model depends on the camera angle and light factors.