SEGMENTASI DAN IDENTIFIKASI KARAKTER PADA PELAT NOMOR KENDARAAN BERMOTOR MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK

• Main Author: Rahmat Mauludin, Tria
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/65205
• Institution: Institut Teknologi Bandung

Automatic Number Plate Recognition (ANPR) is the process to identify a vehicle registration plate. In general, this process consists of three stages. The stages are localizing the vehicle plate, segmentize the character, and identify the character. This final thesis will be focused on developing a model to localize the vehicle plate and a model to identify the character. Both of those models would be developed by using Convolutional Neural Network (CNN). The localize process is approached by the segmentation model using U-Net architecture, the CNN model that was first developed for segmentation in biomedical. The model is built in three different modifications, namely U-Net without modification, U-Net with modification in the number of parameters, and U-Net with VGG19 encoder. The result show that the Intersection over Union (IoU) score for each model are 90%, 86%, and 94%, respectively. For the character identifying process, there are three different classification model that has been developed, namely scratch model and model that created by using transfer learning method, the method that using knowledge from one problem and applying it to related problem. The models that had been applied in that method are VGG16 and VGG19 architecture. The result show that the accuracy for each model are 78%, 86%, and 84%, respectively.