Desarrollo de un sistema de visión artificial basado en redes convolucionales para el correcto posicionamiento de un vehículo autónomo en un carril.

Abstract

The present degree work aims to implement an artificial vision system based on convolutional networks for the correct positioning of a vehicle in a lane composed of two modules. The first, processing images that are captured by an optical sensor through artificial intelligence applying OpenCV, Tensorflow and Keras. The second module is in charge of the control of the motors and the interpretation of the data obtained by the first processing module. The system was subjected to functional tests, determining that the lane detection system is within the precise ranges of interpretation to avoid leaving the limits drawn on a scale lane, the understanding precision of the neural network reaches a 98.54 %. It is concluded that the prototype system allows a stable driving of a vehicle and image interpretation in optimal lighting conditions. For future projects and research, the use of larger image batches and a more robust processing capacity are recommended for more optimal learning and increased autonomous capabilities.