Comparison between Backpropagation and CNN for the Recognition of Traffic Signs

摘要

In this article, it is presented the development of two algorithms of neural networks, which allow the recognition of traffic signs from the realization of a training of neural networks through Backpropagation and another training using neural convolutional networks, in order to compare them and to know which of the two methods is more efficient in terms of execution time and precision, with respect to the same amount of input images that contain the three classes to classify: Regulatory, Warning and Informative. Once the networks were trained, random-image tests were performed achieving 98.33% accuracy for the Backpropagation algorithm and 94.44% accuracy with convolutional neural networks.