The execution of finishing works is usually characterized by deviations from the planned schedule. On the one hand, this is due to the inherent design uncertainties and, on the other hand, due to inevitable disruptions and design changes. For this reason, continuous progress monitoring is crucial to instantly take appropriate actions. Currently, progress monitoring is a very labor intensive, time consuming and subjective process, since on-site data is manually captured and aligned with project information. Accordingly, deviations from schedule are frequently recognized too late, leading to extensive and costly counteractions and design changes. Consequently, robust and efficient construction execution requires real-time, transparent and accurate progress monitoring.

In this research project, innovative and fundamental methods for automated progress monitoring of finishing works using video-based as-built data and 4D as-designed building information models (BIM) are developed. This includes, for example, methods for model-based position and orientation estimation of the camera in order to register videos within the 4D BIM. Consequently, building design information can directly be mapped onto the captured image data. On this basis, machine vision and learning methods are designed to detect and recognize relevant finishing items and their completion states. For this purpose, progress and viewing direction dependent object and state features are defined and stored in an extendable object catalogue. Additionally, novel concepts to store and link relevant position, geometry and material classifiers for finishing items based on the 4D BIM as well as their continuous adaption are the focus of this research. The results are expected to significantly improve the accuracy, the robustness, and the efficiency of progress monitoring for finishing works.