MOURA, V. S.; http://lattes.cnpq.br/4267478145318228; MOURA, Vanessa Santos.
Resumen:
This work aims to contribute with the assistance in the entry and exit of cars in buildings
and private condominiums. Aiming to achieve this goal, this work was divided into two
stages: development of a neural network algorithm capable of identify and recognize the
license plate of a car and development of an algorithm capable of identify and recognize
the driver. For the license plates identification, the pre-trained Tensorflow network model
was used, called MobileNet-SSD, intended for object detection. For the recognition of the
car’s license plate, OCR optical character recognition technology was used. The algorithm
implementation was performed in Python. The deep convolutional neural network showed
good results of precision, recall and F1-score, neural network performance metrics, which
could be improved in future works by performing a pre-processing in the chosen database.
For the identification and facial recognition of the conductor, an oriented gradient histogram
feeding a support vector machine is used. The dlib and face_recognition libraries were
used. Satisfactory results of accuracy, precision, recall and F1-score were obtained, above
98.7%, which demonstrates the predictive fidelity of the implemented facial recognition
algorithm.