Imagine going from a 3D scan point cloud to a standard BIM format such as IFC or CityGML. I know you are skeptical, as am I, but it’s worth a look.
Faramoon is a startup supported by Translating Research at Melbourne (TraM), a program designed to accelerate commercial insight for researchers at the University of Melbourne in Australia. Faramoon is also sponsored by a City of Melbourne (CoM) startup grant to support the integration of the technology to a cloud-based infrastructure to provide this technology outside Australia. The technology is licensed to Faramoon from the University of Melbourne.
They use a five step process beginning with turning unstructured point cloud data to structured. There software produces much better results on structured data. The next step is filtering where the noise and outliers are removed. Methods for filtering including techniques that consider density, distribution and depth of the points in certain clusters, distance between the points, or a combination of these factors.
After filtering the data the third step is recognition. This is where the data is segmented by identifying clusters of points that represent individual building components. There is a range of algorithms and techniques that can be used for segmenting and recognizing components of the buildings. Machine learning can also be used in this step. With this approach, models of building objects to be recognized are stored and labelled in an inventory called the classification database. Then, algorithms that learn are used to recognize new examples of the building objects from the existing objects that are found in the database.
The fourth step is the actual modelling. After the objects are recognized, they are converted into appropriate geometric objects and structured into the required format. Wall thickness needs to be determined from the point cloud of each side of the wall and the required format specifications applied.
The last step is to implement the model. Faramoon has successfully implemented this workflow, by which point clouds of building interiors are automatically processed, and architectural components including ceiling, wall, door, windows and openings are automatically recognized, extracted and converted into semantically-enriched 3D model in two international data standards: CityGML and IFC.
They are claiming a relative discrepancy modelling accuracy between 2 and 5 percent. The only way to know is to try it I suppose.
Read the original article here.