Accurate alignment of heterogeneous LiDAR point clouds is important for producing high-quality elevation data. This process, known as harmonisation, corrects spatial discrepancies between overlapping datasets collected at different times or using different sensors. A key step in harmonisation is the use of reliable benchmarks, which are features that are clearly detectable in both datasets, to guide the alignment. While artificial targets have been traditionally used as such benchmarks, their deployment is costly and impractical for large-scale or uncoordinated surveys. Consequently, there is growing interest in extracting natural or man-made features directly from LiDAR data to serve as benchmarks.

Road markings have recently been proposed as an alternative benchmark due to their consistent visibility in LiDAR point clouds, particularly through intensity value. As part of the Integrale Hoogtevoorziening Nederland (IHN) initiative, road markings are being explored as benchmarks for national point cloud alignment. However, the performance of the road markings as a co-registration benchmark has not been well researched yet.

This research investigates how the accuracy of automatically extracted road markings affects the quality of point cloud alignment. It builds upon an adaptive extraction method that adjusts intensity thresholds to suit different datasets and applies geometric filtering to generate 3D line representations of road markings. These extracted features are used to align heterogeneous LiDAR point clouds, and their performance is evaluated against manually digitised road markings to assess how extraction quality influences alignment accuracy. To support this evaluation, a RANSAC-weighted centroid alignment approach is proposed, which uses the inlier count of each correspondence as a weight during transformation estimation, aiming to prioritise geometrically stable benchmarks in the alignment process.

The results show that alignment accuracy is influenced by the extraction quality and distribution of road markings. When fewer road markings are available, especially in datasets with larger time gaps due to environmental changes, the spatial distribution has a higher chance of becoming uneven, leading to transformation errors. These errors grow with distance from the benchmark area, increasing alignment inaccuracies across the dataset. ... Read more

Accurately classifying laser-scanned point cloud data remains a critical challenge in geospatial analysis, particularly due to the complexity and volume of the data. This thesis presents a novel, confidence-aware deep learning framework designed to improve the classification accuracy of point cloud data, specifically focusing on the Actueel Hoogtebestand Nederland (AHN) dataset. The framework integrates geospatial knowledge into the deep learning process, enabling the model not only to refine its predictions through iterative learning but also to enhance the training data along the way via iterative online learning, ensuring continuous improvement in both training data quality and model performance.

The preprocessing phase assigns confidence scores to each point in the point cloud based on local neighborhood properties, with additional input from multispectral imagery (MSI) to further enhance the confidence estimation. These confidence scores are central to the online learning process, where the model prioritizes high-confidence points for training while progressively updating lower-confidence points to improve accuracy. To test the hypothesis that confidence-aware learning can enhance point cloud classification, we selected the KPConv network due to its suitability for handling unstructured data and capturing complex geometric features.

Extensive experiments demonstrate that the proposed framework, particularly with the Online strategy, enables deep learning models to perform better when trained solely on native point cloud attributes (elevation and intensity) compared to models without this strategy. Importantly, the Online strategy qualitatively enhances the training data by refining labels and reducing noise, thereby supporting more robust model performance. While incorporating additional features from aerial imagery showed no overall improvement, specific classes, like High tension and others did see performance gains. ... Read more