LULC - Special topic

Land Use and Land Cover III

6607.3 - A clustering-based segmentation method for mobile laser scanning data

Thursday, July 6
3:30 PM - 3:50 PM
Location: Maryland B

With the developing capability for acquiring high density point cloud data, mobile laser scanning systems has been widely utilized for various applications, such as 3D city modeling, road inventory study, safety control, car navigation, forestry management etc. Data from mobile laser scanning systems (MLS), airborne laser scanning systems (ALS) and terrestrial laser scanning systems (TLS) are all being applied in urban scene analysis. However, MLS data is more suitable for urban scene information extraction for two reasons. First, compared to ALS data, MLS data has higher density and contains more vertical information, which are of great importance in identifying information from poles and buildings. Moreover, MLS data is more efficiently acquired than TLS data, as the latter one is collected by moving systems. Information extracted from MLS data, such as road, light poles and building information, are usually added to a GIS (Geographic Information System) database as a new layer.
Region grow(Vo et al. 2015), model-based(Aijazi et al. 2013; Vosselman 2013) and graph cut(Golovinskiy et al. 2009; Yang and Zhen 2013) based method are commonly used methods that can segment MLS data effectively in generally simple cases, however, they suffer from the following limitations. First, region grow methods are computationally expensive for multi-dimensional features in large datasets with iterative computation and they cannot detect consistently data points around edges. Second, since these algorithms only use point positions, many spurious planes that do not exist in reality may be generated in model-based methods. Third, for graph cut methods the results was strongly based the energy threshold, while with uneven densities no such global threshold exist.
To overcome the aforementioned limitations, we present a new clustering based method to segment mobile laser scanning data. First, the unorganized point clouds are voxelized to compress and re-organize the original data. Then the ground voxels are extracted and filtered out from the original voxel set based on some criterion. Third, the density based clustering algorithm is performed to segment the filtered voxel set where street objects are over-segmented in this stage. Finally, the voxel set is back-projected to point cloud to merge the over-segmented objects and also a major filtering method is applied to deal with those unassigned points left in the clustering stage.
The performance of the proposed method is validated by two datasets. The experiments indicate the completeness values in the test sites is over 90%, while the correctness values vary differently with the datasets and are from 85% to 96%. Mistaking street objects mostly occurs when objects are partially reached by laser beams due to being sheltered by buildings or are staying too close to each other.
The main advantages of the proposed method are as follows: (1) it needs no additional information except for the coordinates and intensity values of the point clouds; (2) it uses the bottom-up tracing algorithm to calculate density, which overcomes the uneven-density problems for the density-based method; (3) the proposed method can even segment trees and poles out in a nested situation where trees and poles are interlaced with each other. The preliminary experimental results show that the proposed method can efficiently segment street objects in urban scenes.

You Li

Wuhan University, College of Resource and Environmental Science

A doctor candidate in College of Resource and Environmental Science, Wuhan University, who is major in Geographic Information System. His research topic is feature extraction from Mobile Laser Scanning data.

Presentation(s):

Send Email for You Li

William a. Mackaness

Senior Lecturer in GIS
The University of Edinburgh

William Mackaness is a senior lecturer at The University of Edinburgh in the School of GeoSciences. His research is in map generalisation - developing automated techniques for display of geographic information at multiple levels of detail. His interest extends into mapping via smartphone technologies. His work in Malawi and with aboriginal communities in Australia has led to an interest in the socio technical dimensions of GIS.

Presentation(s):

Send Email for William Mackaness


Assets

6607.3 - A clustering-based segmentation method for mobile laser scanning data



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for A clustering-based segmentation method for mobile laser scanning data