Number plate recognition on vehicle using YOLO - Darknet
Budi Setiyono, Dyah Ayu Amini, Dwi Ratna Sulistyaningrum
Abstract
Abstract Character recognition is one of the steps in the number plate recognition system. Character recognition is done to get text character data. The method used is YOLOv3 (You Only Look Once), and Darknet-53 is used as a feature extractor. In this study, the data used were number plate images derived from the extraction and cropping of motorized vehicle videos that had been taken using cellphones and cameras. Testing is done with two different models, namely the model obtained with additional preprocessing data and the model obtained without any preprocessing data. Data preprocessing is done to improve the quality of the number plate image. Testing is done on an uninterrupted number plate image dataset and a number plate image dataset with interference and a reduction of color intensity (brightness) in the image. For the model obtained from the data without preprocessing, the highest number plate recognition accuracy obtained is 80%, and the character recognition accuracy is 97.1%. Meanwhile, for the model obtained from preprocessing data, the highest number plate recognition accuracy obtained was 88%, and the character recognition accuracy was 98.2%.