Automatic License Plate Recognition for Parking System using Convolutional Neural Networks
Joshua Joshua, Janson Hendryli, Dyah Erny Herwindiati
Abstract
Traditionally, a parking staff enters the license plate of the outgoing vehicle in the system to generate the parking tickets. To improve the efficiency of the system, we propose an automatic vehicle license plate numbers recognition to automatically detect and record the plate numbers to the system eliminating the manual entry. The system consists of a YOLO model to automatically detect the license plate from the image of a vehicle, some preprocessing and image segmentation process to detect the digits in the license plate, and finally, a ResNet model to classify the plate numbers. The license plates used in this research are from Indonesia. From the experiments, we find that the YOLO can accurately detect the license plate with a high degree of confidence. Yet, the ResNet model achieves around 80% accuracy from the validation data. Despite the high accuracy, the model can sometimes wrongly classify plate numbers due to noises from the segmentation process, non-standard or damaged plates, and similarly looking digits.