Automatische herkenning van netwerken uit puntenwolken [Engels]

Organisatie || Siers Groep
Afstudeerder || Jorn Kruiper (B.Sc. thesis)

Update: Februari 2024

In most situations, contractors should provide a digital drawing of all cables and pipes. Currently, new underground utilities are mapped by a land surveyor.  New technologies such as LiDAR and photogrammetry can make 3D point clouds of utility trenches. Extracting or segmenting the utilities from a point cloud can be done using machine learning. The goal of machine learning is to produce an algorithm that can learn patterns present in a dataset to perform a specified task. This research aims to compare different machine learning algorithms for automatically recognising and retrieving underground utilities in 3D point clouds of open trenches and converting the data into geometric shapes.

Unlabelled Point Cloud

First, point clouds of utilities are pre-processed and labelled in two categories: ‘utility’ and ‘not utility’. In the figure above the original point cloud is presented. This point cloud is further partitioned in a labelled point cloud (blue lines) that visualizes utilities.

Labelled Point Cloud containing Utility Lines

The data is then fed to three different machine learning algorithms that were identified as the best options. The models are trained on a training dataset. During training, the models are tested on a test dataset. Assessment criteria are calculated from the outcomes of the models. These assessment criteria are used to compare the different models. The best model is used for a test case so that the outcome of the model can be visualised. Finally, the utilities are converted to geometric shapes that can be used for mapping.

Point Cloud Containing Only Utilities used for Training