Prototipo de asistencia visual para detectar objetos en espacios de aprendizaje aplicando redes neuronales convolucionales

Abstract

Throughout this technical project, a visual assistance prototype containing sensor modules to distinguish distances from obstacles around an educational environment has been developed, the main purpose of which has focused on applying a convolutional neural network. This initial prototype enables object recognition and facilitates the use of an auditory user interface that allows listening through headphones to the detected objects. For the creation of the neural network, the Edge impulse platform was used which, based on the creation of a dataset with objects referring to the educational classroom, made it possible to configure the image, parameters, generate characteristics, adjust the convolutional neural network, train the model, visualize the metrics and, finally, the network that was built made it easier to work with Arduino. The results of the development of this prototype by applying functional tests were carried out with several students and were effective demonstrating in each use that it has a different fluency with different objects while the accuracy depends on some factors such as the background and the light, however it was demonstrated that it correctly fulfils the function and the objectives of the project.