Enhanced Object Detection Methods for Streamlined Inspection Process in Automotive Applications

Abstract

This project presents an end-to-end desktop application for automatic manhole detection using a twostage deep-learning pipeline wrapped in a user-friendly Tkinter GUI. The system begins by ingesting adirectory of street-level images and corresponding YOLO-format annotations, which are normalized,shuffled, and split into training and testing sets. A lightweight convolutional neural network (CNN) isfirst trained— or loaded