Reality capture for construction quality

Models describe what should be built. Reality capture records what is actually there. Laser scanners, photogrammetry from drones and handheld cameras, 360-degree site cameras and mobile mapping systems can capture existing conditions, progress and installed work in hours, producing point clouds, meshes and imagery that can be compared with the design. CON-250 explained how scanners and drones measure and the rules for flying drones. This lesson focuses on how a construction manager uses captured data for quality assurance: specifying accuracy and density, scan-to-BIM for existing buildings, scan-versus-model comparison of installed work, progress monitoring, and managing the large data sets involved.

Capture methods compared

Each capture method trades accuracy, speed, cost and coverage differently. Terrestrial laser scanners on tripods give the highest accuracy over building-scale distances. Mobile and handheld scanners, often using simultaneous localization and mapping (SLAM), are much faster but less accurate and can drift over long paths. Photogrammetry builds 3D geometry from overlapping photographs; from drones it covers large sites quickly, while from the ground it records façades and interiors. 360-degree cameras, often worn on a helmet during a site walk, produce navigable photo records rather than precise geometry.

Choosing a method starts with the question being asked. Checking that embedded anchor bolts are within a few millimeters needs a total station or a terrestrial scanner, not a 360-degree camera. Documenting what is inside a wall before drywall goes on needs a fast, complete photo record, not millimeter accuracy. Many projects use several methods together.

Specifying accuracy and density

Two separate properties describe a point cloud. Accuracy is how close the measured points are to their true positions, usually expressed as a standard deviation or a tolerance range at a confidence level. Density is how closely spaced the points are on the surface, expressed as point spacing (for example, one point every 6 mm at 10 m) or points per square meter. A cloud can be dense but inaccurate, or accurate but too sparse to show small features.

The U.S. Institute of Building Documentation publishes a Level of Accuracy (LOA) specification that defines accuracy bands for measured data and for represented data (models made from it). For example, LOA 30 corresponds to a range of about 5/8 in down to 3/16 in (15 mm to 5 mm) at two standard deviations, and LOA 40 to about 3/16 in down to 1/16 in (5 mm to 1 mm). A scope of work should state the LOA required for measurement and for the resulting model, by element type if necessary, together with the control network and registration method.

Registration, the process of aligning separate scans into one coordinate system, introduces its own error. Scans can be registered to each other by matching overlapping geometry (cloud-to-cloud) or by targets, and tied to the project coordinate system through targets surveyed with a total station from site control (CON-250). For quality assurance comparisons against a model, tying scans to project control is essential; otherwise any registration error looks like a construction error.

Scan-to-BIM for existing buildings

Renovations, additions and retrofits begin with existing conditions that are often poorly documented. Original drawings may be missing or wrong, and decades of alterations rarely appear on them. Scan-to-BIM means capturing the existing building and modeling it from the point cloud, so the design team starts from reality rather than from assumptions. It is now routine on hospital renovations, heritage buildings, industrial plants and complex interiors.

Scan-to-BIM should be scoped carefully. Modeling every pipe and cable tray in an existing ceiling space at high accuracy can cost more than the design itself. The scope should say which elements to model, to what LOD and LOA, and which areas can stay as point cloud only for reference. Designers increasingly work directly with the point cloud as a background in their authoring tools, modeling only what they need to change or connect to.

Scan versus model: checking installed work

On new construction, scanning installed work and comparing it with the coordinated model is a powerful quality check. Deviation analysis computes the distance from each point to the nearest model surface and colors the result: green within tolerance, yellow near the limit, red beyond it. Typical uses include checking slab flatness and elevations before finishes, verifying embeds and sleeves before a pour, confirming that MEP installation matches the coordinated model before ceilings close, and checking structural steel or precast placement.

The following script simulates deviations between scan points and a design surface, then reports the statistics a quality manager would look at.

import random, statistics

random.seed(8)
# Simulated deviations (mm) of 2,000 scan points from the design slab surface:
# most points near zero, plus a sagging bay near one column.
devs = [random.gauss(0.5, 3.0) for _ in range(1850)] + [random.gauss(-12, 3.0) for _ in range(150)]

tol = 8.0   # +/- 8 mm tolerance for this check (illustrative)
within = sum(abs(d) <= tol for d in devs)
print(f"points: {len(devs)}")
print(f"mean {statistics.mean(devs):+.1f} mm, std dev {statistics.stdev(devs):.1f} mm")
print(f"min {min(devs):+.1f} mm, max {max(devs):+.1f} mm")
print(f"within +/-{tol:.0f} mm: {within} ({within / len(devs):.1%})")
low = sum(d < -tol for d in devs)
print(f"below tolerance: {low} points -> inspect for sag or low spots")

points: 2000
mean -0.4 mm, std dev 4.4 mm
min -20.6 mm, max +10.0 mm
within +/-8 mm: 1844 (92.2%)
below tolerance: 142 points -> inspect for sag or low spots

The summary statistics show most of the slab within tolerance but a cluster of low points. In practice the color map would show where they are, and the team would decide whether to grind, fill or accept them, depending on the specified tolerance and the finish to be applied. Concrete floor flatness and levelness are normally specified with their own measurement methods, and a scan comparison supplements rather than replaces the specified test.

Progress monitoring

Repeated capture builds a history of the site. Weekly 360-degree walks or drone flights let the project team, the owner and remote specialists see the site as it was on any date. Comparing successive captures with the 4D model (Lesson 6) shows which elements are installed, which are late and which are ahead. Some platforms use computer vision to detect installed elements automatically, such as studs, drywall, ductwork or concrete, and report percent complete by area.

Automated progress detection is improving quickly but is not perfect. Occlusion (elements hidden behind others), similar-looking elements and partial installation confuse the algorithms. A sensible approach is to use automated results as a first pass, verified by the superintendent, rather than as an unquestioned measure for payment.

Documenting concealed work

One of the most practical uses of reality capture is documenting work before it is concealed. A 360-degree walk or scan of every wall and ceiling before drywall or ceilings close creates a record of what is inside: pipes, conduit, blocking, fire stopping, insulation, vapor barrier. Months or years later, a maintenance technician can see where a pipe runs before cutting into a wall, and a dispute about whether fire stopping was installed can be settled from the images. Owners increasingly require this documentation at handover.

Managing reality capture data

Point clouds are large. A single high-resolution terrestrial scan can be hundreds of megabytes; a full building can run to hundreds of gigabytes or more. Weekly captures over a two-year project multiply that. The BIM execution plan should say where data will be stored, in what formats (E57 is the common open format), at what resolution for each use, who can access it and how long it will be kept. Decimated or tiled versions are used for viewing, with full-resolution data archived.

Privacy also matters. Site cameras and 360-degree walks capture images of workers, neighbors and the public. Projects should tell workers how images are used, avoid capturing private property where possible, blur faces and license plates where required, and follow privacy law (Lesson 11). Drone flights must follow aviation rules and any site-specific restrictions (CON-250).