Print Email Facebook Twitter De-noising terrestrial laser scanning data for roughness characterization of rock surfaces Title De-noising terrestrial laser scanning data for roughness characterization of rock surfaces Author Altunda?, D. Contributor Khoshelham, K. (mentor) Nygan-Tillard, D. (mentor) Menenti, M. (mentor) Faculty Aerospace Engineering Department Earth Observation and Space Systems Programme Geomatics Date 2009-06-23 Abstract To measure roughness a number of traditional measurement methods have been developed such as linear profiling, compass and disc clinometers. Although these measurement methods have been widely used, they are time consuming, and labor intensive. The accuracy of the local measurements depends heavily on expertise and measurements are limited to places that are accessible. Therefore, the demand for ways of acquiring roughness information quickly and with a high spatial resolution and accuracy for surfaces out of human reach is increasing. Terrestrial laser scanning (TLS) is a data acquisition technique promising to fill all these needs. However, data acquired by terrestrial laser scanner devices always contain noise. Therefore, using the raw laser data for roughness characterization can cause inaccurate and unreliable results. In order to confirm that terrestrial laser scanning technique can replace the traditional measurement methods for roughness characterization it is essential to remove noise present in the data as accurate as possible without distorting the detailed features underlying the original data. The research presented in this thesis focuses on analyzing the influence of terrestrial laser scanner (TLS) instrument noise on rock surface roughness characterization. To characterize roughness the roughness length method is used. Linear profiles are generated on the rock surface and from these profiles roughness data are obtained using both terrestrial laser scanner data and manual measurements. The manual measurements are used as reference for the evaluation of the results from the laser scanner data. From the profiles the fractal parameters are obtained to quantify roughness using the roughness length method. Both raw data and denoised laser scanner data are used to obtain fractal parameter values. With respect to the selection of the most appropriate threshold value to distinguish the original data from the laser range noise, an experiment was performed using the wavelet denoising approach. In this experiment, the threshold values were changed incrementally until the fractal dimension values approached the values of the reference data. If the selected threshold value was too small to remove noise then the denoising procedure resulted in the noise being underestimated (e.g. 0.001m for our case). In contrast, if the threshold value is too high then the denoising procedure resulted in the noise being overestimated. The results clearly showed that the selection of threshold values in denoising procedure play a significant role in changing the values of the fractal dimension. However this method was only used to see the effect of the threshold value on the fractal parameters. It was concluded that the best approach is to first estimate the noise level in the laser data and then determine the threshold values. For this reason two methods were applied to determine the noise level (Donoho-Johnstone’s and Menenti-Ritchie’s methods), and then they are used in different threshold estimation methods. In total 10 different thresholding methods, including the hard and soft modes, were performed on the wavelet coefficients obtained separately from discrete wavelet transform and wavelet packets. Donoho-Johnstone’s, Fixed form, hard thresholding method and Birge-Massart’s, Penalized medium soft, thresholding methods when applied to the coefficients of wavelet packets resulted in very close fractal dimension values to the reference values in x directional laser profile. However, for y directional profiles Birge-Massart’s Penalized medium hard thresholding method resulted in the closest value to the reference data. In this research it is showed that noise can be successfully eliminated to such a degree that accurate measurement values can be obtained from laser scanned data for rock surface roughness characterization. Subject terrestrial laser scanningde-noisingroughnesswavelet To reference this document use: http://resolver.tudelft.nl/uuid:219f6326-95c9-445a-aa87-413cc80ea819 Part of collection Student theses Document type master thesis Rights (c) 2009 Altunda?, D. Files PDF Denoising_Terrestrial_Las ... rfaces.pdf 7.5 MB Close viewer /islandora/object/uuid:219f6326-95c9-445a-aa87-413cc80ea819/datastream/OBJ/view