Construction site monitoring and as-built BIM using photogrammetry

As-design and as-built Building Information Models have gained increased popularity in the Architecture, Engineering and Construction industry. The number of countries that require such models during the constructional workflow and at project completion has increased significantly in the past years. Apart from documentation purposes, it is hoped that this demand will also propel innovation in the construction industry, commonly known as rather conservative. Even today, in a very digitised and automated world, the construction workflow with in particular the monitoring of executed works by site managers, mainly happens in a manual way. In the eternal drive to optimise profits, this fact forms a strange anomaly. In this project a series of (semi-)automated workflows is developed to augment the current construction site monitoring practices. The performed research follows a logical start-to-end pipeline, analogue to the manual monitoring analyses in-place, but in a more digitised and data-driven fashion. The following contributions are part of this research.

Data acquisition
Construction site images form the main input of all developed frameworks. However, only since recently the information they contain can be harnessed fully. Factors that have led to this include the advent of digital cameras that accurately depict the building scene, the availability of sufficient processing power and the increasingly performant photogrammetric pipelines. Nevertheless, construction environments remain very challenging to capture and pose many obstacles. This research covers the devices, methodology and challenges to accurately record image data. Current data acquisition workflows are examined and extended towards construction industry purposes to facilitate accurate, yet accessible data recording sessions.

Data processing
Only if the captured imagery can be positioned correctly (relative to each other but also to other datasets such as the as-design BIM model), the information the pictures contain can be successfully used in data-driven monitoring approaches. Relying on the traditional photogrammetric pipelines and advancing the state-of-the-art, a framework is developed that realises to process and (geo-)reference multi-temporal image data without the need for tedious and error-prone ground control point indication. One of the major conclusions is that, using the presented approach, not only valuable time can be saved but also the accuracy of the construction site datasets rises.

Data analyses
Once the imagery has been photogrammetrically processed, it can be used to compare the reality to the design. The detected discrepancies between both worlds allow for updating the as-design to an as-built BIM model. Two separate frameworks are developed for these analyses. A first relies on the 3D geometry of the point cloud data to determine all element deviations. By not only considering the optimal shift of the considered element itself but also the dominant transformation of the cluster of its nearest neighbours, deviations can be determined virtually unaffected by georeferencing and drift errors. While we prove this to be true for the performed experiments with synthetic data, optimisations are still required to transfer the method’s potential to realistic environments. Secondly, a purely image-based approach was developed that evaluates element positions by comparing reality images with duplicate virtual images of the BIM environment, called BIMages. The latter are created via the known characteristics of all images retrieved in the preceding photogrammetric process. The fact that reality image and BIMage do not fully agree in the case of element deviations is exploited. Displacements are intentionally induced to the BIM element and through image pair similarity evaluations it is assessed what the optimal displacement is, i.e. what the deviation of that element is. Excellent results are achieved in the experiments showcasing the method’s large potential. However, tests in more complex building environments are still required.

Data visualisation
While experts are able to detect various anomaly patterns, this frequently is not the case for site managers who are less familiar with three-dimensional remote sensing data and deviation analyses. Therefore, the final step consists of insightfully outputting the obtained results via clear and comprehensible visualisations or via an even more immersive Virtual Reality environment. The developed proof of concept applications allow for timely presentations of recorded data and analysis results, thus satisfying an industry where developments follow each other rapidly and timely error detections are crucial to lower the omni-present failure costs.