Machine learning approaches for the uncertainty modelling of a low-cost laser scanner

In many applications, low-cost laser scanners (also LiDAR) are used to capture the environment. Common applications of these sensors are the localization of robots in logistics halls or of autonomous cars. Further applications include mobile mapping, allowing for the rapid capturing of large volumes. Although the sensors are widely used, their uncertainty properties have not yet been thoroughly investigated. Especially in comparison to high-end laser scanners, which are often used for geodetic measurements.

The aim of this master thesis is to develop a quality model for a low-cost laser scanner. In this case, a Velodyne VLP-16 would be used. The modelling of the uncertainty should include different features that influence the measurement (distance, material of the scanned object, ...). The influence of these features should be quantified. The modelling should employ methods from machine learning. In addition, methods based on parametric forward uncertainty modelling (with expert knowledge) will be used as a basis for comparison of the machine learning model.

At the end, the developed models should be validated with real data. For this purpose, the measurements of a low-cost scanner should be compared with the measurements of a geodetic laser scanner.

Language: English/German

Recommended previous knowledge: MATLAB / Python, principles of laser scanning/LiDAR systems