Real-time pixel-level fusion of colour and thermal (RGB-T) imagery via extrinsic optical alignment: Scope for automating building inspections

Abstract

Building inspections and energy surveys are performed to detect and diagnose building defects, however, existing inspection methods are highly labour intensive, time consuming, and not suitable for large-scale audits. This work presents a thermal-colour fusion technique using pre-capture optical alignment that eliminates the need for compute-intensive post-processing of data. The coinciding axes of dual cameras also prevent parallax errors, which helps expand its applications to near proximity or indoor thermal inspection as well. A prototype of this design was assembled and tested, which showed a pixel-level accurate alignment of thermal and colour camera imagery with minimal computation. The research and development in this paper provides scope and feasibility to automate building inspection techniques through drone-based low-powered embedded machine learning models and identify faults in real-time for large-scale surveys of building façades.