Exploring Different Options of Imbalanced Learning for Outlier Detection in Point Cloud Data

Mapping underwater areas, such as riverbeds, are the first steps of underwater exploration. Advancements of different methodologies such as range scanning methodologies, both optical and acoustic techniques, has provided the possibility to get high resolution information of these areas in the form of 3D point clouds. These high-density observations are also contaminated with noise and outliers.

This Master thesis is part of WaMUT project. The data are results of UAV and USV measurements. The aim is to mathematically model the point cloud as a continuous surface. To achieve this goal the first step is to preprocess the data and detect outliers. In the framework of this master thesis different approaches based on machine learning will be adopted to find the most effective process for detecting outliers with these data characteristics. At the end a mathematical model for the 3D point cloud is to be derived based on B-splines surfaces.

Language: English

Recommended previous knowledge: MATLAB / Python, Basic Knowledge of statistics