Condition Monitoring of Road Traffic Infrastructure by Using Synthetic Aperture Radar

Kurzfassung

Roads play a crucial role in the development of a country. Therefore, it is essential to carry out periodic inspections to assess the road surface conditions and carry out necessary maintenance activities for ensuring the efficient and safe movement of people and goods. The most important factors affecting the quality of road surfaces include surface roughness, cracks, potholes as well as unevenness. Nowadays, specialised survey vehicles equipped with numerous sensors are used globally to monitor road conditions, however, this activity is mainly concentrated on major roads and only once in every few years, as this is a very resource-intensive task. However, as road conditions deteriorate rapidly, particularly in winter due to freeze-thaw cycles, more frequent monitoring is essential to detect problems early and take preventative actions. The utilisation of airborne and spaceborne synthetic aperture radar (SAR) systems is a promising avenue for road condition monitoring as it offers cost-effective and large-scale monitoring capabilities, enabling predictive road maintenance strategies. This doctoral thesis focuses on the development of methods and processing chains for accurate road surface roughness estimation, detection and orientation estimation of cracks as well as road width estimation using high-resolution fully focused SAR data. The road surface roughness estimation methods were developed and tested for both airborne and spaceborne X-band SAR systems, while the processing chains for crack detection, orientation estimation and road width estimation were specifically tailored and tested for airborne X-band SAR systems. A new semi-empirical roughness estimation model and machine learning-based support vector regression (SVR), random forest regression (RFR) and artificial neural network (ANN)-based regression models were developed, trained and tested specifically for road surface roughness estimation using fully polarimetric airborne X-band SAR data. The new semi-empirical roughness estimation model and processing chain were adapted for road surface roughness estimation using single-polarised X-band spaceborne SAR data. Since both the airborne and spaceborne SAR data have a low signal-to-noise ratio (SNR) due to the lower backscatter from smooth road surfaces, additive noise estimation and minimisation techniques were integrated into the processing chains as a pre-processing step to improve the reliability of the road surface roughness estimation. Similarly, after generating the road surface roughness values, upper sigma nought and lower SNR thresholding techniques were implemented to further eliminate the invalid and noisy results. Multi-dataset fusion approaches were also developed to fuse the surface roughness estimates from multiple SAR datasets with different data acquisition geometries to minimise the errors introduced in road surface roughness estimation due to incidence angle variations, low SNR and shadow regions. Furthermore, a novel method based on the combined use of an adaptive thresholding algorithm and the Radon transform has been proposed in this thesis for cracks detection, severity and orientation estimation. In this method, the cracks detection is performed using the adaptive threshold algorithm, while the severity of the cracks is expressed in terms of the maximum Radon magnitude values obtained from the sinogram and the orientation of the detected cracks is represented as bearing angles. The road width is measured by detecting abrupt changes in road surface roughness at the road boundaries. All processing chains, including those for road surface roughness, cracks detection, orientation estimation and road width measurements, generate keyhole markup language (KML) files to visualise the results in Google Earth (GE) for enhanced interpretation. Validation of these methods and processing chains was conducted using DLR's airborne F-SAR system and Germany's spaceborne TerraSAR-X system, demonstrating close agreement with ground truth measurements, which were acquired with sub-millimetre level accuracy by using a 3D laser scanner. Overall, this research contributes to advancing road condition monitoring techniques, with implications for predictive maintenance strategies and infrastructure management.