Metrological evaluation of an AI-based vision computing model for crack detection on masonry structures

Ensuring the structural integrity of buildings is essential for their longevity and safety. Traditional methods of surface monitoring, crucial for detecting potential damages that could lead to structural failures, are often labour-intensive, subjective, and challenging to document comprehensively. This paper proposes an innovative, automated approach to address these challenges by leveraging advanced computer vision and artificial intelligence. The method focuses on the detection of cracks in masonry building elements, a common but critical indicator of building surface wear. Utilizing a robust AI model trained on a diverse dataset of real crack images, the crack area is identified, and the system is able to accurately determine crack dimensions, encompassing both width and length, by analysing the contour of this area. An analysis was carried out on synthetically generated images to determine which parameters most significantly affect the detection capabilities of the AI model, and validation of real crack images was performed. Our approach redefines building monitoring by combining the precision of machine learning and vision systems techniques with the strategic insights provided by a comprehensive platform, setting a new standard for structural health management in the construction industry