D4ar- 4 dimensional augmented reality - models or automation and interactive visualization of construction progress monitoring

Early detection of actual or potential performance deviations in field construction activities is critical to project management as it provides an opportunity to initiate proactive actions to avoid these deviations or minimize their impacts. Despite the importance, (1) current monitoring methods require manual as-built data collection and extensive as-planned data extraction; (2) due to extensive workload, observations are sometimes conducted infrequently and progress is measured with non-systematic metrics; and (3) current reporting techniques are visually complex which requires more time to be spent on communicating the status of a project. There is a need for a systematic approach allowing data to be collected easily, processing the information automatically and reporting back in a format useful for all project participants. This research addresses these challenges by introducing D4AR – 4D Augmented Reality – models as integrated as-built and as-planned environments. These models, generated for automated tracking and visualization of construction performance deviations, take advantage of two emerging sources of information: (1) Unordered daily construction photo collections, which are nowadays collected at almost no cost on all construction sites; and (2) Building Information Models (BIMs), which are increasingly turning into binding components of Architecture/Engineering/Construction (AEC) contracts and if linked with construction schedule, can serve as powerful baselines for tracking and visualization of performance deviations. In this research, an approach based on structure-from-motion technique is presented which operates on a set of unordered and uncalibrated daily construction photographs, automatically computes photographer's locations and orientations, and generates a 3D point cloud representation of the as-built site. Within such an environment, images are registered in 3D, allowing large unstructured collections of daily photos to be sorted, interactively browsed and explored. Reconstructed as-built point clouds, generated with different photo collections assembled in different days, are automatically superimposed over one another using an iterative closest point algorithm and consequently result in 4D as-built models. Next, 4D BIMs are fused into 4D as-built point cloud models by control based registration-steps and generate D4AR models. The as-built point cloud models are enhanced with a multi-view stereo algorithm and are fed into a novel voxel coloring and labeling algorithm to increase density of the as-built point cloud models, and traverse and label the integrated as-built and as-planned models for expected visibility and observed occupancy. Finally, a machine learning scheme built upon a Bayesian probabilistic model is presented which automatically detects physical progress in presence of occlusions and demonstrates that component-based progress monitoring at schedule activity-level could be automated. The system developed in this research enables as-planned and as-built models to be jointly explored with an interactive, image-based 3D viewer wherein deviations are automatically color-coded over the BIM using a simple traffic-light metaphor. The resulting D4AR models overcome the challenges of current progress monitoring practice and further enable AEC professionals to conduct various decision-making tasks in virtual environments rather than the real world where it is time-consuming and costly. To that extent, the underlying hypotheses and algorithms for generation of integrated 4D as-built and as-planned models as well as automated progress monitoring are presented. Promising experimental results are demonstrated on several challenging building construction datasets under different lighting conditions and sever occlusions. This marks the D 4AR modeling approach to be the first of its kind to take advantage of existingconstruction photo collections for the purpose of automated monitoring and visualization of performance deviations. Unlike other methods that focus on application of laser scanners or time-lapse photography, this approach is able to use existing information without adding burden of explicit data collection on project management and reports competitive accuracies compared to those reported with laser scanners especially in presence of sever occlusions.