Automatic number plate recognition via convolutional neural network for residential gate access control

The traditional guardhouse visitor management system at residential gate that practises manual checking of residents' and visitors' passes can result in traffic congestion during peak hours. To address this issue, this study proposes an Automatic Number Plate Recognition (ANPR) system using a customised Convolutional Neural Network (CNN) to automate the residential and visitor verification process by recognising the number plate, thereby reducing traffic congestion. In addition to the existing CNN-based ANPR system, this study investigated the performance of the combination of computer vision techniques with a custom CNN image classification model. Comparison analysis was carried out between YOLOv3 and computer vision methods, and between MobileNetV2 and a custom CNN model to identify the most effective techniques for number plate localisation and character recognition. The models' performance was evaluated using 150 images, where the custom CNN model outperformed MobileNetV2 with an accuracy of 0.997. Image augmentation was introduced to diversify the training set, where the custom CNN model with augmented data achieved an accuracy of 0.998 and an F1 score of 0.999. The results suggest that the proposed CNN-based ANPR system has the potential to automate the residential verification process and reduce traffic congestion.