Document Type

Theses, Ph.D

Rights

This item is available under a Creative Commons License for non-commercial use only

Disciplines

Remote sensing

Publication Details

Successfully submitted for the award of Doctor of Philosophy to the Dublin Institute of Technology 2010.

Abstract

Airborne sensors including LiDAR and digital cameras are now used extensively for capturing topographical information as these are often more economical and efficient as compared to the traditional photogrammetric and land surveying techniques. Data captured using airborne sensors can be used to extract 3D information important for, inter alia, city modelling, land use classification and urban planning. According to the EU noise directive (2002/49/EC), the National Road Authority (NRA) in Ireland is responsible for generating noise models for all roads which are used by more than 8,000 vehicles per day. Accordingly, the NRA has to cover approximately 4,000 km of road, 500m on each side. These noise models have to be updated every 5 years. Important inputs to noise model are digital terrain model (DTM), 3D building data, road width, road centre line, ground surface type and noise barriers. The objective of this research was to extract these objects and topographical information using nationally available datasets acquired from the Ordnance Survey of Ireland (OSI). The OSI uses ALS50-II LiDAR and ADS40 digital sensors for capturing ground information. Both sensors rely on direct georeferencing, minimizing the need for ground control points. Before exploiting the complementary nature of both datasets for information extraction, their planimetric and vertical accuracies were evaluated using independent ground control points. A new method was also developed for registration in case of any mismatch. DSMs from LiDAR and aerial images were used to find common points to determine the parameters of 2D conformal transformation. The developed method was also evaluated by the EuroSDR in a project which involved a number of partners. These measures were taken to ensure that the inputs to the noise model were of acceptable accuracy as recommended in the report (Assessment of Exposure to Noise, 2006) by the European Working Group. A combination of image classification techniques was used to extract information by the fusion of LiDAR and aerial images. The developed method has two phases, viz. object classification and object reconstruction. Buildings and vegetation were classified based on Normalized Difference Vegetation Index (NDVI) and a normalized digital surface model (nDSM). Holes in building segments were filled by object-oriented multiresolution segmentation. Vegetation that remained amongst buildings was classified using cues obtained from LiDAR. The short comings there in were overcome by developing an additional classification cue using multiple returns. The building extents were extracted and assigned a single height value generated from LiDAR nDSM. The extracted height was verified against the ground truth data acquired using terrestrial survey techniques. Vegetation was further classified into three categories, viz. trees, hedges and tree clusters based on shape parameter (for hedges) and distance from neighbouring trees (for clusters). The ground was classified into three surface types i.e. roads and parking area, exposed surface and grass. This was done using LiDAR intensity, NDVI and nDSM. Mobile Laser Scanning (MLS) data was used to extract walls and purpose built noise barriers, since these objects were not extractable from the available airborne sensor data. Principal Component Analysis (PCA) was used to filter points belonging to such objects. A line was then fitted to these points using robust least square fitting. The developed object extraction method was tested objectively in two independent areas namely the Test Area-1 and the Test Area-2. The results were thoroughly investigated by three different accuracy assessment methods using the OSI vector data. The acceptance of any developed method for commercial applications requires completeness and correctness values of 85% and 70% respectively. Accuracy measures obtained using the developed method of object extraction recommend its applicability for noise modelling

Share

COinS