Development of a Computer Vision Algorithm Based on a Convolutional Network to Detect Mortar Joints According to the National Building Regulation (RNE) and Automate Site Verifications

León León R.A.; Llanos Meza A.A.; Marroquín Zavaleta M.A.; Reyes Gonzales T.R.; Tejada Villegas G.D. · CISCI - Conferencia Iberoamericana en Sistemas, Cibernetica e Informatica (2024) · DOI: 10.54808/cisci2024.01.447

Resumen

In the construction industry, manual inspection of mortar joints poses significant challenges due to human errors and lack of clear criteria. To address this, an artificial vision algorithm based on YOLOv8 was developed to detect mortar joints in accordance with the National Building Regulations (RNE), thereby automating technical inspections on-site. 550 mortar joint images were collected and labeled using Roboflow, diversifying lighting conditions and angles. YOLOv8 was utilized in Google Colab with a Tesla T4 GPU for model training, parameter adjustment, and validation with independent datasets. Implementation was done in PyCharm for real-time detection. The model achieved an average mortar joint detection accuracy of 92%. It was observed that 68%-75% of joints were separated by 1-1.5 cm, and 25%-32% by more than 1.5 cm. These results highlight the system's effectiveness in enhancing precision and efficiency in construction inspection. © CISCI - Simposium Iberoamericano en Educacion, Cibernetica e Informatica 2024.

Instituciones: Universidad Privada del Norte