Deep learning for multi-attribute vehicle recognition in vehicle access control system

Wong, Yuan Zhen (2025) Deep learning for multi-attribute vehicle recognition in vehicle access control system. Final Year Project, UTAR.

PDF Download (12Mb)

Abstract

Vehicle recognition systems are becoming increasingly essential for intelligent transportation, traffic surveillance, and security applications. For the purpose to identify vehicle attributes in real-time, this research provides a hybrid vehicle recognition system that is implemented as a web and mobile application. It combines deep learning and computer vision techniques. To extract license plates, colors, makes, models, and production years of vehicles, the system mainly uses EasyOCR and YOLOv8 with multi-attribute detection. A refined GPT-4o visual-language model (VLM) acts as a fallback, improving recognition reliability in edge instances when YOLOv8 and EasyOCR would not yield reliable results. The approach involves capturing pictures of vehicles from cameras or user uploads, processing them using the pipeline for detection and OCR, then using the GPT-4o VLM to verify the outcomes. The system's modular design provides seamless connection with web and mobile platforms, enabling real-time performance and scalability. It is expected to work reliably across a variety of lighting situations, angles, and occlusions. The study shows a potential approach for enhancing the recognition of vehicle attributes. Future research will involve adding more unusual vehicle kinds to the dataset, refining the model inference for edge devices, and integrating predictive analytics for anomaly detection and vehicle tracking. Keywords: vehicle recognition; YOLOv8; EasyOCR; GPT-4o; deep learning; computer vision; license plate recognition Subject Area: T58.5-58.64 Information technology

Actions (login required)