The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume XLII-2/W15
https://doi.org/10.5194/isprs-archives-XLII-2-W15-497-2019
https://doi.org/10.5194/isprs-archives-XLII-2-W15-497-2019
22 Aug 2019
 | 22 Aug 2019

ALGORITHMS FOR THE AUTOMATIC DETECTION AND CHARACTERIZATION OF PATHOLOGIES IN HERITAGE ELEMENTS FROM THERMOGRAPHIC IMAGES

I. Garrido, S. Lagüela, S. Sfarra, and M. Solla

Keywords: Heritage, conservation, moisture, InfraRed Thermography, automation, image processing

Abstract. Heritage elements, from historic buildings to stone sculptures and panels, stand as key elements in the history of humanity. Unfortunately, the deterioration of both the surface and the interior of these elements is inevitable, endangering the quality and existence of these structures of high historical value in the event of a delay in the implementation of the required maintenance tasks. InfraRed Thermography, IRT, appears as one of the most recent techniques to detect and characterize possible pathologies in structures in their early stages, being very useful for a preventive analysis in heritage elements.

This paper presents a methodology for the automatic detection and characterization of one of the most severe and frequent pathologies in heritage structures, moisture, from thermal images. The proposal stands as a demonstration of the potential of the IRT technique for heritage conservation applications, and as a new step towards the automation of the inspection process and optimization of the decision-taking in conservation actions within cultural heritage. For that, two thermal criteria and a semi-automatic image rectification process are implemented as main phases of the methodology, obtaining good results for the detection of moisture zones and accurate area values with regard to the real dimensions of each moisture zone. Specifically, an F-score average of 78 % ± 19 % regarding detection performance and a percentage relative error of minimum 4 %, and maximum of 12 %, referred to the area computation in unit metrics are obtained.