Sistema de detecção e identificação automática de placas de trânsito brasileiras aplicada ao auxílio à direção veicular utilizando técnicas de processamento digital de imagens

Abstract

The Advanced Driver Assistance Systems (ADAS) are systems that seek to help the driver to drive the vehicle safely, either passively or actively. The ADAS has several characteristics, however, the vehicle's ability to map the surrounding environment stands out, identifying hypotheses of risk for the driver, passengers, or pedestrians, and must act effectively or alert about such conditions. One of the applications of this environment mapping refers to its usability in detecting and identifying traffic signs. For this process to be carried out with an adequate level of precision, the use of computer vision techniques, digital image processing and, more recently, Neural Networks are used. Thus, this work proposes to develop a system for the detection and automatic identification of Brazilian traffic sign plates, applied to aid in vehicular driving, to use digital image processing techniques. The model consists of a RCNN - (Region Based Convolutional Neural Networks) system, consisting of two stages: detection and classification. In the first, using image processing techniques and SVM model to extract the zone of interest, obtaining a rate accuracy of 67.2% in a test environment, testing with benchmark dataset. The second step consists of using the result of the previous process to carry out the classification of traffic signals, following a classic Convolutional Neural Network model, reaching an accuracy rate of 99.26%, in a test with benchmark dataset, and for database of Brazilian plates, 85.29%, being considered satisfactory for the dataset.