Clasificación de vehículos y reconocimiento de placas usando técnicas de visión artificial.

Abstract

Abstract:Currently, there is evidence of an influx of traffic in the urban areas of Ecuador, since each year the automotive fleet of the cities increases considerably and generates data or information that can contribute to the entities in charge of traffic management. In this curricular integration work, an object detector was developed which allows classifying the different types of vehicles most common in certain sectors. For the construction of the prototype, frameworks such as TensorFlow and PyTorch were used. In addition, a total of six neural networks were trained and evaluated, three of them for vehicle classification and the others for license plate detection and recognition. The final results determined that the YOLOv7 model used in the prototype intended for vehicle classification had accuracy scores for SUV 94%, SED 96%, CMT 91%, BUS 80% and CML 88% with respect to the license plate detection evaluation. an accuracy of 75% was obtained, finally, the optical character recognition was positive with a character error rate of less than 10% in most of the tests.