Automatic indoor construction progress monitoring: challenges and solutions

Abstract: Indoor construction progress monitoring has challenges like occlusion, light variation, dynamic environment which makes its automation different from outdoor construction progress monitoring. AI/deep learning approaches can help overcome these challenges but using them for indoor construction monitoring raises some issues like lack of annotated image data for construction works. Transfer learning provides the initial solution to the AI related challenges. In our research we use this state-of-the-art method on construction site data and detect as-built stages of a drywall construction. The results are promising with accurate prediction of 3 stages of the drywall process.