Through 3D scanning technology, Cognite obtains accurate point clouds of large-scale industrial plants with a high level of detail. We're currently using these point clouds for visualization and measurement purposes to provide value to our customers.

However, we're still missing a fundamental understanding of the objects located in a three dimensional scene. This understanding starts with semantic segmentation, the process of assigning each a class to each point in a point cloud. In the D-MVP and 3D team, we are currently working hard on solving this with state-of-the-art Deep Learning technology. Our goal is to automatically connect assets in a point cloud directly to Cognite Data Fusion, enabling rapid development of fully contextualized as-built digital twins.

In the attached video you'll see what we are aiming to achieve: A point cloud from an industrial site, where all the points are split into different classes. Let me know what you think eller I'd love to hear your thoughts