Abstract
This research investigated the use of deep learning and computer vision techniques for the continuous detection of cracks in concrete buildings, aiming to enhance structural health monitoring and preventive maintenance. The study utilised a quantitative and explanatory approach using a dataset of 4,000 labelled images of cracked and non-cracked surfaces from Kaggle. Data pre-processing and model training were done using Python-based tools such as Pandas, OpenCV and TensorFlow, with a convolutional neural network (CNN) developed to classify images into faulty and non-faulty categories. The model achieved an accuracy of 84%, with precision, recall and F1-score values indicating strong predictive performance, though a moderate recall for cracked surfaces demonstrates the need for improved detection sensitivity.
To enhance real-world applicability, the study further explored the Internet of Things (IoT) integration, proposing an edge-based monitoring system using devices such as Raspberry Pi and NVIDIA Jetson Nano for real-time crack detection. The findings showed Deep Learning offers a scalable, efficient and automated option to conventional inspection and non-destructive testing methods. The research concluded that a combination of CNN-based computer vision with IoT-enabled systems can enable continuous, autonomous and cost-effective infrastructure monitoring, aiding in safer and smarter building maintenance practices.