Automated parking and payment system using license plate and vehicle attribute recognition with multimodal ai models

Yong, Ting Wei (2025) Automated parking and payment system using license plate and vehicle attribute recognition with multimodal ai models. Final Year Project, UTAR.

PDF Download (7Mb)

Abstract

As of October 2023, Malaysia recorded over 36.3 million registered vehicles, highlighting the need for more efficient and intelligent parking solutions. Traditional parking systems, which rely on physical tickets, RFID tags, and e-wallets, often lead to congestion, delays, and security vulnerabilities. This project proposes an AI-powered parking payment system that integrates License Plate Recognition (LPR) with Vehicle Attribute Recognition using the Gemini 2.5 Flash multimodal large language model (LLM), complemented by a mobile application designed for drivers, operators, and administrators. The recognition module, developed in Python, was tested using a ground truth dataset of 20 real vehicle images from Roboflow in a simulated environment. Each image was passed directly to the Gemini model to extract license plates and vehicle attributes such as make, model, and color. Recognition was incorporated into two system points: (1) a mobile app feature allowing users to auto-fill vehicle details via photo uploads, and (2) simulated parking facility entry and exit points where vehicle identity was verified against a backend database for automated payment processing. The mobile app was written with Laravel, React Native, and PostgreSQL, and it offers role-based features including vehicle registration, payment tracking, and operational oversight. Testing showed an 85% accuracy for full multi-attribute recognition, with individual accuracies of 95% for license plates, 100% for color and make, and 90% for model detection. Average recognition processing time was 2.495 seconds per image upload. While entry and exit recognition were simulated, the system successfully demonstrated automated vehicle verification and payment workflows. The mobile application facilitated seamless user interactions and system management. Limitations include reliance on free-tier AI services, absence of real-time hardware integration, and limited analytics capabilities. This project illustrates the feasibility of leveraging multimodal LLMs and mobile platforms to create ticketless, contactless, and fraud-resistant parking solutions, contributing to Malaysia’s digital transformation and smart city initiatives. Keywords: artificial intelligence; license plate recognition; vehicle attribute recognition; smart parking; fraud prevention; multimodal LLMs; smart city Subject Area: QA75.5–76.95 Electronic computers. Computer science

Actions (login required)