Countries currently have large road networks that require a huge amount of traffic signs. Overhauling and maintaining all roads and keeping an uptodate inventory of their traffic signs is hard and timeconsuming. Therefore, it is necessary to automate the recognition of traffic plates in order to guarantee better efficiency and costeffectiveness for the maintenance of signs. In this work, a deep learning approach is adopted, using an artificial neural network for the detection and recognition of traffic signs on roads. The model used was the Mask RCNN, capable of being trained endtoend and performing the instance segmentation. An own dataset was created to focus on Brazilian roads, as the publicly available datasets use signs from their respective countries, making their use unfeasible. The trained model has excellent precision (mAP0.5 = 80.95 percent), considering the dataset size and the limited time and computational resources. Such a system has shown enormous potential for its commercial use in managing and maintaining an updated inventory of traffic signs.