Vguard: Development of mobile real-time vehicle damage detection application with image recognition using YOLOV8

ABSTRACT: Traditionally, vehicle inspections rely on manual procedures that often lack precision, efficiency, and visual feedback. To address these limitations, this study developed VGuard, a mobile application that enhances vehicle diagnostics by integrating automated engine data retrieval, real-time image recognition, and visual representation of system status and faults. VGuard highlights three core features: (1) real-time engine data retrieval using an ELM327 OBD-II scanner diagnostic trouble codes and sensor data from the Engine Control Unit (ECU); (2) exterior damage detection using YOLOv8, trained on a dataset of 4,609 labeled images of dents, scratches, cracks, and rust; and (3) dual visualization, which includes a 2D fault display and a 3D car model for better interpretation of both internal issues. The trained YOLOv8 model achieved a mean Average Precision (mAP) of 72.1%, precision of 82.7%, and recall of 63.7%, allowing the system to recognize vehicle surface damages accurately. However, the model showed a higher rate of false positives for scratches and background elements, which may be attributed to poor lighting, image noise, and limited training diversity. The integration of OBD-II diagnosis, intelligent image recognition, and informative visual displays improved the usability of the system and made vehicle condition assessment more accessible. Overall, VGuard successfully met its development goals and presents a promising mobile solution for real-time vehicle damage detection and diagnosis.

Filipiniana