Desarrollo de un sistema de visión artificial basado en redes convolucionales para la detección de las señales de tránsito implementado sobre un vehículo autónomo.

Abstract

This paper describes the analysis, development and creation of an artificial vision system based on convolutional networks for the detection of traffic signs, speed control and detention, implemented on an autonomous vehicle, for which the processes that are carried out in advanced driver assistance systems, also the algorithms of artificial vision system were studied. The prototype bases its operation on the use of free software by programming in Python with OpenCV libraries and the training of the convolutional network for pattern recognition using techniques of artificial vision such as the Haar Cascade algorithms, with the use of these the processing necessary for the detection of traffic signals and the control of vehicle speed. On the recognition of these signals, a Raspberry OV5647 camera was used, which was placed on the top of the vehicle, so the module received the information and processed it through the algorithm using the training described giving the control order on the driver L298N, which controls the vehicle's travel speed. Based on the tests carried out, it was determined that the detection of signals has a 99.16% accuracy; therefore, it is concluded that the implemented prototype presents a high degree of functionality at the time to meet its objective, being able to be improved and applied to real vehicles, being capable of contributing to the reduction of traffic accidents. It is recommended to use 8GB Raspberry pi 4B due to its efficiency in processing; if you need more resources, you can use accelerators designed for artificial vision applications to obtain greater precision and fewer mistakes in recognition.