Sensitivity analysis of unmanned aerial vehicle-borne 3D point cloud reconstruction from infrared images

Abstract

Structural aging or inefficient design affects the energy efficiency of buildings in the United States. Infrared thermography (IRT) in conjunction with unmanned aerial vehicle (UAV) followed by point cloud (PC) reconstruction can be used to generate renderings of structures that highlight areas of excessive heat loss. This work represents the first effort to standardize the procedures for developing a structure from motion (SfM) approach by using only IR images to generate three-dimensional (3D) PC models (i.e., renderings) that can be used to determine from where heat is lost in a structure. Additionally, the appropriate image acquisition parameters are poorly understood and are investigated within this paper to improve the reconstruction of infrared-based point clouds (IR-PCs) for buildings. In particular, the sensitivity of the IR-PC approach for reconstructing 3D models and its accuracy to detect heat loss is quantified as a function of the i) temperature difference between the targeted structure and its surroundings, ii) IR images' overlap, and iii) IR camera's orientation. Results of tests performed to demonstrate that the IR-PC allows reconstructing virtual models of the targeted systems with geometrical accuracy of about 2 cm and for temperature differences as low as 1.3 °C. Overall, this research defines a standard for using only IR images to reconstruct 3D PCs that can be used to detect heat loss by determining the optimal image acquisition and flight plan parameters. This framework can find applications for structural assessment and building inspection but also make energy efficiency assessments more precise and much more widespread than they currently are.