Spectral Filtering and Classification of Terrestrial Laser Scanner Point Clouds |
| |
| Authors: | Derek D Lichti |
| |
| Institution: | () Curtin University of Technology, Perth, Western Australia |
| |
| Abstract: | A method is presented for filtering and classification of terrestrial laser scanner point clouds. The algorithm exploits the four-channel (blue, green, red and near infrared) multispectral imaging capability of some terrestrial scanners using supervised, parametric classification to assign thematic class labels to all scan cloud points. Its principal advantage is that it is a completely data-driven algorithm and is independent of spatial sampling resolution since the processing is performed in four-dimensional spectral feature space. Its application to two data-sets of different spatial extent and spatial and spectral complexity is reported, for which respective overall classification accuracies of 87·0% and 82·0% were achieved. Analysis of the input data with emphasis on the characteristics pertinent to the anticipated outcomes precedes detailed analysis of the classification results and error sources and their causes. Erroneously classified points are attributed to radiometric errors stemming from both detector hardware and physical effects. |
| |
| Keywords: | multispectral classification point cloud terrestrial laser scanning |
|
|
