Development of a robot vision system for detection of wall cracks

Abstract

The advancement of Artificial Intelligence (AI) has revolutionised machine learning and has been widely implemented in many applications. The aim of this project is to utilise Convolutional Neural Network (CNN) to develop an inspection system for detection of wall cracks in newly constructed buildings. This report details the process of implementing the real-time detection system on the live video footage streamed by an automated flight path-planning drone. It entails the documentations of training the YOLOv3 wall crack detection model, from the data collection stage, fine-tuning of hyperparameters, and finally, integrating the detection model with the real-time footage captured by the drone to accomplish real-time detection and localisation of wall cracks. Comprehensive and detailed analysis of the wall cracks are then obtained, and engineers on the ground may use these to assess the structure of the buildings and make the necessary rectifications.