Processing Point Clouds for Infrastructure: Why Building Standards Aren't Enough

Scanning a commercial office building is relatively forgiving. Scanning a three-kilometer highway bridge or a massive dam is entirely different. Infrastructure-grade BIM demands a level of data precision that far exceeds typical commercial standards. Here’s how we ensure the data is actually usable for heavy civil engineering.

The problem with standard scanning

When you scan a room, the software can easily match the flat floors and 90-degree walls together. But civil infrastructure is full of repetitive, featureless geometry: endless stretches of identical bridge deck or perfectly smooth tunnel linings.

If you just run standard "cloud-to-cloud" auto-registration on a highway, the software gets confused. It slides the scans together incorrectly, and suddenly your bridge model is curving into the sky. You cannot rely on the software's best guess when you are designing structural retrofits.

Nailing the registration

We follow a rigid registration hierarchy for infrastructure. We never rely exclusively on cloud-to-cloud matching.

Instead, our survey crews physically place survey targets—checkerboards or spheres—across the site, tied to a highly accurate GNSS control network. Back in the office, we force the software to register the scans using those specific physical targets first. We only use cloud-to-cloud algorithms as a secondary refinement step to tighten things up. For an LOD 350 deliverable, we demand a registration error of less than 3 millimeters between any two scan positions.

Sorting the data

A raw point cloud is just a massive billion-point mess of color. To make it useful for BIM, we have to classify it.

Standard commercial scanning schemas don't work for infrastructure. We've had to extend the standard classification systems to handle civil assets. Our teams strip out the construction noise and temporary props, then segment the data into specific structural classes: the main structural decks, the bridge piers and abutments, the safety barriers, and the hidden drainage weeping holes.

Moving to BIM (IFC Compliance)

The goal isn't a pretty point cloud; the goal is an engineered BIM model. When we extract geometric primitives—like a concrete pier—from the point cloud data, we ensure the final deliverables conform to IFC 4.x standards.

Every element we model carries metadata: the exact survey origin coordinates, the allowable scanning tolerance, the date the scan was taken, and a flag indicating if the structure has moved or changed since the last baseline scan.

QA isn't optional

Before we ever hand a point cloud over to a design team, it has to pass three internal gates. First, the registration report has to explicitly prove the error tolerances were met. Second, the automated scripts check that no required classified layers are missing. Finally, we pull 50 random check coordinates from the finished BIM model and measure them directly against the raw laser data to prove the model actually reflects reality. In heavy infrastructure, a 2-inch deviation matters.

Building an infrastructure digital twin? Get in Touch →