CAR PLATE DETECTION USING REGION-BASED CONVOLUTIONAL NEURAL NETWORKS
Abstract
Convolutional Neural Networks (CNN) comprise a deep learning technology which is widely used to perform image classification. In this research, we review CNN structures and explain how they can be used for finding license plates in vehicle images. We summarize how the standard CNN processes images into features and compare it to Region-Based Convolutional Neural Networks (R-CNN). After comparing their pros and cons, we decide to design a R-CNN to train our dataset for this project. We find that the one trained with the most training data has the highest testing accuracy. The first training network detector leads to the highest testing accuracy that can reach 0.963 after 10 training epochs. The second training network detector leads to the highest testing accuracy that can reach 0.988 after 20 training epochs.