This paper presents a method for Brazilian automatic license plate recognition for Smart Cities applications, considering images with good illumination and alignment conditions, with lack of illumination, with shadows and with some inclination, taken from a distance of lm to 7m. Images were obtained in a parking lot, during the morning and the afternoon, without environment restrictions. License plate extraction is performed by a cascade of Haar-like features classifiers, built using the AdaBoost algorithm. Filters based on mathematical morphology prepare candidate plates for binarization and a simple threshold rule segments them. Characters are extracted by connected components labelling and are recognized by template matching. Results showed 81.75% of accuracy with average processing times of 210ms and 1304ms for computers with quad-core processors in frequencies 2.2GHz (Dell Inspiron 5588) and 900MHz (Raspberry Pi 2 model B), respectively.