Sensitivity analysis of full-waveform LiDAR data for material classification

Full-waveform (FWF) LiDAR can provide additional features which can be helpful to distinguish between different construction materials. However, due to its high volume FWF information is often discarded and not used for downstream analysis, such as material classification. This contribution investigates waveform processing, such as Gaussian decomposition, and modelling to extract important radiometric features from terrestrial laser scanning. In addition, geometric information is extracted by analysing point cloud neighbourhoods. The resulting combination of features is evaluated by means of sensitivity analysis to obtain their corresponding relevance to material classification. Specifically, the assessment is performed leveraging machine learning algorithms, such as support vector machines, and monitoring the influence of the model’s performance depending on the combinatoric inclusion of the proposed individual features. To support the analysis, furthermore a rich dataset is presented, consisting of point clouds, waveform and image data of various urban and infrastructure scenes. Thereby the aim of the classification problem is to semantically segment the point clouds according to common materials such as concrete surfaces, vegetation or brick. This effort serves the goal of improving the automatic digitization of construction assets through the use of advanced remote sensing techniques.