# A Real-Time Mobile Robotic System for Crack Detection in Construction Using Two-Stage Deep Learning

## TL;DR

This paper introduces a robotic system that autonomously detects cracks in construction using deep learning, enabling real-time inspections with high accuracy.

## Contribution

A novel two-stage deep learning framework combining U-Net and Pix2Pix GAN for improved crack detection accuracy in real-time robotic systems.

## Key findings

- The two-stage model achieved a mean Intersection over Union (mIoU) of 73.9% and an F1-score of 76.4% on the CrackSeg9k dataset.
- The system successfully detected micro-cracks as narrow as 0.3 mm in real-world campus hallway tests.
- The robotic system enables simultaneous crack detection and autonomous navigation using an RGB-D camera and LiDAR.

## Abstract

The deterioration of civil infrastructure poses a significant threat to public safety, yet conventional manual inspections remain subjective, labor-intensive, and constrained by accessibility. To address these challenges, this paper presents a real-time robotic inspection system that integrates deep learning perception and autonomous navigation. The proposed framework employs a two-stage neural network: a U-Net for initial segmentation followed by a Pix2Pix conditional generative adversarial network (GAN) that utilizes adversarial residual learning to refine boundary accuracy and suppress false positives. When deployed on an Unmanned Ground Vehicle (UGV) equipped with an RGB-D camera and LiDAR, this framework enables simultaneous automated crack detection and collision-free autonomous navigation. Evaluated on the CrackSeg9k dataset, the two-stage model achieved a mean Intersection over Union (mIoU) of 73.9 ± 0.6% and an F1-score of 76.4 ± 0.3%. Beyond benchmark testing, the robotic system was further validated through simulation, laboratory experiments, and real-world campus hallway tests, successfully detecting micro-cracks as narrow as 0.3 mm. Collectively, these results demonstrate the system’s potential for robust, autonomous, and field-deployable infrastructure inspection.