Fugro
Fugro has collected huge amounts of data relating to the Earth’s surface and subsurface. Manual methods of classification are time-consuming and expensive. In the last year, we have developed algorithms using Julia to provide per-point classification of 3D point clouds using our own spatial feature libraries in combination with classifiers such as XGBoost. This work has been so successful that we’ve been able to provide detail about a range of real-world objects, such as buildings, roads, powerlines, vehicles and fences, to our clients without requiring expensive and slow human quality assurance processes. In this talk I’ll discuss the machine learning algorithms we use, approaches for optimisation, data storage, and our greatest challenges and opportunities going forward.
Josh is a data scientist/engineer at Fugro, where he develops machine learning algorithms to model the world from lidar and imagery data.
