Estudo de IA para reconhecimento de objetos e cores aplicada a sinais de trânsito

Abstract

The project aims to recognize traffic lights in images available in a folder with a pre-trained model and extracts from these images every object it understands as a traffic light. These images are subsequently followed and separated into “red traffic light”, “yellow traffic light”, “green traffic light” and what was recognized as a traffic light but is not a traffic light in the “negative” folder. After separating these images, it is used to train the InceptionV3 network to recognize and differentiate traffic lights by color, a part of this data is separated for testing and another for network validation. The highest performing convolutional neural network and saved in an file. This file is used in a code to recognize and classify traffic lights in colors in images captured in real time. In summary, an AI trained to recognize and classify traffic lights by color can solve several problems and challenges related to safety, efficiency and reliability when applied to autonomous vehicles or in conjunction with an ADAS system, providing safe interaction with the environment. traffic, adding the benefit of continuous AI training and learning.