Image-based construction of building energy models using computer vision

• Published in: Automation in construction Vol. 116; p. 103231
• Main Authors: Dino, Ipek Gursel, Sari, Alp Eren, Iseri, Orcun Koral, Akin, Sahin, Kalfaoglu, Esat, Erdogan, Bilge, Kalkan, Sinan, Alatan, A. Aydin
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.08.2020; Elsevier BV
• Subjects: 3D modeling; 3D reconstruction; Building energy modeling; Buildings; Computer simulation; Computer vision; Construction; External walls; Ground truth; Image quality; Image reconstruction; Infrared imaging; Performance measurement; Thermography; Three dimensional models; Building energy modeling; 3D modeling; Computer vision; 3D reconstruction; Infrared imaging
• ISSN: 0926-5805; 1872-7891
• DOI: 10.1016/j.autcon.2020.103231

Improving existing buildings' energy performance requires energy models that accurately represent the building. Computer vision methods, particularly image-based 3D reconstruction, can effectively support the creation of 3D building models. In this paper, we present an image-based 3D reconstruction pipeline that supports the semi-automated modeling of existing buildings. We developed two methods for the robust estimation of the building planes from a 3D point cloud that (i) independently estimate each plane and (ii) impose a perpendicularity constraint to plane estimation. We also estimate external walls' thermal transmittance values using an infrared thermography-based method, with the surface temperatures measured by a thermal camera. We validate our approach (i) by testing the pipeline's ability in constructing accurate surface models subject to different image sets with varying sizes and levels of image quality, and (ii) through a comparative analysis between the calculated energy performance metrics of a ground truth and calculated energy simulation model. •Image-based 3D reconstruction pipeline for semi-automated building energy modeling•A thermography-based method implemented to calculate actual thermal resistance•Two methods proposed for plane segmentation of 3D point clouds•Comparative analyses conducted with different sets of images in a classroom•Constrained segmentation method more robust with low-quality, less number of images