Integration of Large Language Models into Computer Vision Based Traffic Monitoring and Vehicle Classification

Proponent/s Name/s (Last Name, First Name, Middle Initial)

• Reece Eiro M. Abuy, De La Salle University, ManilaFollow
• Jaezar Marcellius F. Navalta, De La Salle University, ManilaFollow
• John Zane L. Mirabueno, De La Salle University, ManilaFollow
• Jai Andrei D. Villanueva, De La Salle University, ManilaFollow
• Immanuel Jose C. Valencia, De La Salle University, ManilaFollow

Document Types

Paper Presentation

Research Theme (for Paper Presentation and Poster Presentation submissions only)

Computer and Software Technology, and Robotics (CSR)

School Name

De La Salle University

Track or Strand

Science, Technology, Engineering, and Mathematics (STEM)

Research Advisor (Last Name, First Name, Middle Initial)

Valencia, Immanuel Jose C.

Start Date

25-6-2026 1:30 PM

End Date

25-6-2026 3:00 PM

Zoom Link/ Room Assignment

Online - https://zoom.us/j/91936856247?pwd=oCMfMsh44I2wb0dYsEgoInDJy59bOq.1 Meeting ID: 919 3685 6247 | Passcode: research

Abstract/Executive Summary

Urban traffic congestion in the Philippines represents a critical crisis, with Metro Manila commuters losing an average of 257 hours annually to delays (Subingsubing, 2020). This study aims to develop an AI-based traffic monitoring system that integrates YOLO-based computer vision for vehicle classification with Large Language Models (LLMs) to transform technical traffic data into human-readable insights. Using a curated dataset of 8,000 images from Taft Avenue, the methodology involves training object detection models to recognize a diverse mix of local vehicles including jeepneys and tricycles and processing metrics like Volume-to-Capacity Ratios (VCR) through a GPT-4 pipeline. Results demonstrate that the Yolo26-L and Yolo26-X variants achieved a peak mean Average Precision (mAP50) of 99.5%, confirming high technical reliability for real-time analysis. The research concludes that this integrated framework provides a scalable, objective, and data-driven solution for urban planning, effectively bridging the gap between raw detection metrics and actionable decision-making for smart city development in the Philippines.

Keywords

computer vision; vehicle classification; traffic congestion; LLMs; AI

Initial Consent for Publication

yes

Statement of Originality

yes