Engineers at process plants and related facilities such as pipelines and tank farms often need to create 3D and geographic information system (GIS) models of their installations. These models can be used for planning new construction, augmenting HMIs and mapping existing facilities. 3D models also have a place in improved operator training through simulation and other techniques.
In the past, these models were created by either manual manipulation of digital photos or with light detection and ranging (LIDAR). But now a technique called 3D reconstruction from imagery is automating many of the manual steps required to create 3D models and offers some advantages over other techniques.
Joe Seppi is the practice leader for national security at Woolpert, a design, geospatial and infrastructure management firm, and he's used all three methods to create 3D models.
"The advantages of 3D reconstruction from imagery over more traditional approaches to 3D modeling are speed, the visual quality of the fully textured meshes and colorized point clouds. These meshes and clouds are virtually impossible to make using the traditional photogrammetric workflows needed to turn digital photos into 3D models. The disadvantages are in accuracy, precision and the ability to generate attributed vector feature classes that are GIS-ready," explains Seppi.
"LIDAR is intrinsically a very precise survey tool, and LIDAR point clouds can be used to generate high-quality 3D models. LIDAR is also an active sensor technology with some advantages over 3D reconstruction in terms of discriminating between elevated features and ground, resulting in more precision of the 3D model. Although generating vectorized 3D models from LIDAR point clouds is a common practice, a great deal of additional work is needed to texturize LIDAR-derived models with imagery," notes Seppi.
"3D reconstruction software solutions such as Acute3D, recently acquired from Bentley, can turn digital photographs from any camera into 3D meshes," claims Pascal Martinez, Bentley's senior product manager for Acute3D.
"To model every aspect of a given building or scene, the goal in the image acquisition process is to take a minimum of three sharp overlapping photographs. Anyone following this rule will be able to capture suitable imagery datasets—whether using a smartphone, compact DSLR or a high-resolution professional camera," continues Martinez.
The next step is turning these digital photos into usable 3D models. "First, the Acute3D software automatically locates the position and orientation of each photograph and then estimates the optical properties of the camera. Next, the software extracts and refines a 3D reality mesh that is texturized directly from the photos. This 3D reality mesh can be produced in various formats to interoperate with third-party applications such as CAD and GIS software," points out Martinez.
Seppi and Woolpert use Acute3D and are mostly pleased with the software. "What is exciting about full 3D reconstruction and Acute3D is the level of automation and, consequently, the quality of results that can be achieved in a relatively short period of time," says Seppi.
As with any new software product, Acute3D has benefited from feedback from Seppi and other early adopters. "In early runs of Acute3D we saw anomalies around water and thin painted lines on paved surfaces that would take hundreds of hours to clean up manually. Each time we brought these to the attention of Acute3D support, they did a good job of issuing patches and new releases to eliminate the problems. There will always be parts of the model that can benefit from some touching up, and for that we follow Acute3D's suggested third-party workflow for touching up mesh geometry and textures," concludes Seppi.