Modelling discontinuous terrain from DSMs using segment labelling, outlier removal and thin-plate splines

Kassel Liam Hingee, Peter Caccetta, Louis Caccetta

Research output: Contribution to journalArticlepeer-review

3 Citations (Scopus)


Models of ground surface elevations are crucial to many applications of remotely sensed data, including estimates of the height relative to ground of non-ground objects, such as buildings and vegetation. In highly engineered regions, such as cities, there are many discontinuities in both the ground surface and the surface of non-ground objects. Sub-metre resolution elevation data for these regions are increasingly available. At these resolutions there is sufficient information and a growing need to improve model accuracies by incorporating discontinuities. Here we provide a new method for generating high resolution models of discontinuous ground surfaces from breakline data and digital surface models derived from remotely sensed data. The method uses segment based filtering, outlier removal and multiresolution thin-plate spline surface fitting. Breaklines are included in the fitted surface using partial derivatives and a breakline-aware method for transferring between different resolutions. We demonstrate our method using elevation data derived from photogrammetry for suburban regions of Perth, Western Australia, and Vaihingen, Germany. We produced ground surface models with noticeable qualitative and quantitative improvements when breaklines are included, at an increased computational cost of approximately 10% when all other parameters remained the same. For LiDAR derived elevations, we report our residual error against a number of other methods recorded using the ISPRS Ground Filtering Test Sites.

Original languageEnglish
Pages (from-to)159-171
Number of pages13
JournalISPRS Journal of Photogrammetry and Remote Sensing
Publication statusPublished - 1 Sep 2019


Dive into the research topics of 'Modelling discontinuous terrain from DSMs using segment labelling, outlier removal and thin-plate splines'. Together they form a unique fingerprint.

Cite this