Motorcycle License Plate and Driver Face Verification Using Siamese Neural Network Model

Abstract

The security and efficiency of vehicle access management systems have become a primary concern for various institutions, including universities, offices, and public facilities. Effective access management not only enhances security but also improves the flow of incoming and outgoing vehicles, reduces congestion, and enhances user experience. This research aims to develop a vehicle plate detection system and driver face recognition using the Siamese Neural Network model to optimize traffic at the gate. The methods used include the application of deep learning algorithms, specifically the Siamese Neural Network, to verify the driver's face and the use of You Only Live Once (YOLO) to detect and recognize vehicle plates in real-time. Data was collected through direct capture with the researcher's camera. The model was trained and tested using a dataset containing images of vehicle license plates and driver faces. The results showed that the developed model was able to detect and recognize the vehicle plate and the driver's face with a fairly high accuracy, namely in the object detection results getting bounding box validation is 1.05 and class loss validation is 0.95, and 0.85 mAP. As well as in training using the Siamese Neural Network, the highest result is 0.82 with a learning rate of 10e-5 with 30 epochs. It is hoped that this system can be one of the innovations that can be applied in government agencies, universities, industries, etc.