Large-scale range data acquisition and segmentation

The research reported in this thesis aims to devise a cost-effective, robust technology and technique for accurately measuring and segmenting geometric details embedded in the exterior surfaces of large buildings. A diverse range of applications become viable or significantly enhanced by capturing the accurate geometric data of significant buildings and extracting the fine details of such data. Motivations for this thesis stem from the facts that commercially available large-scale data acquisition systems are expensive and existing algorithms for 3D data processing are yet to address the processing complexity associated with the 3D segmentation of large outdoor objects. The contributions of this thesis are threefold. First, a low-cost versatile large-scale rangescanner, capable of capturing range data up to 300 metres, is designed and implemented. An innovative method for system calibration and the data fusion of the intensity and range measurements has been developed. A number of experiments are also conducted to evaluate the performance of the proposed rangescanner system and the data fusion technique. The range data obtained by the rangescanner system has been verified using two methods of verification - application checking and equivalence checking. Example results of the verification, presented in this thesis, shows that the laser rangescanner device is capable of providing adequate accuracy and resolution for large-scale civil application with minimum complexity and cost. The design provides portability, flexibility and ease of operation. Secondly, problems associated with range data acquisition and processing of large building exteriors are studied. Key challenges of processing the range data of large buildings, including significant disparities in the size and depth and the existence of substantial construction error in historical buildings, have been identified and their effects for the segmentation task are examined. Thirdly, a computationally effective and robust segmentation technique, capable of extracting geometric details of large building exteriors, is developed. The segmentation algorithm, titled Hierarchical Robust Segmentation (HRS), uses a high breakdown, robust estimator in a hierarchical coarse-to-fine approach. This algorithm is then tested on several range data sets obtained by different laser rangescanners. The experimental results show that the proposed algorithm overcomes most of the current problems of large-scale data segmentation presented in this thesis by extracting both coarse and fine details from range data of large building exteriors in a relatively short period of time.