From registration to recognition of indoor construction states using on-site videos and 4D building models

Interior finishing works cover a big budget part of every construction project. Current practice is lacking procedures and methods for frequent onsite progress monitoring, so that unexpected and unrecognized delays during this... [ view full abstract ]

Interior finishing works cover a big budget part of every construction project. Current practice is lacking procedures and methods for frequent onsite progress monitoring, so that unexpected and unrecognized delays during this phase of construction result in a decreased overall project performance.

To address this problem, this paper synthesis previously presented parts of a method that recognizes the actual state of construction activities from as-built video data and as-designed 4D BIM data using computer vision algorithms. Under this method, two main steps are required: First, the video frames are registered with the underlying 4D building model, meaning that the camera position and orientation of each video frame is determined within the coordinate system of the building model. During this step an iterative registration procedure is proposed that combines relative and absolute pose estimation. Second, once the camera poses are known, the relevant construction activities and elements from the 4D building model are projected onto the image space to determine regions of interest, which are then taken as input for computer vision based activity state recognition methods. 

Compared to previously presented conceptual work, this paper puts a strong emphasis on experiments and test results. As the overall method consists of several consecutive steps, each single process is first tested individually, before the combined procedure and the general applicability of the method is evaluated by means of two exemplary types of construction activities. All experiments show very promising results and reveal the method’s potential to support automatic indoor construction progress monitoring.