Terrestrial Laser Scanning Data Analysis for Deformation Monitoring

Deformation monitoring of structures is one of the main tasks of engineering geodesy. In the projects related with deformation monitoring, terrestrial laser scanning (TLS) has become a powerful tool among all the data acquisition approaches due to its high precision and spatial resolution in capturing 3D point clouds. Therefore, it allows to entirely monitor the behavior of objects. The
challenge lies in the definition and computation of the difference between the 3D point clouds in order to model the deformations. Various methods exist, from which the geometry-based method is one of the most popular ones. The key procedure in this strategy is to approximate the point cloud of an epoch by mathematic models, mostly in a linear Gauss-Markov model. The geometry
changes of the object are detected by comparing the approximated models of various epochs. The traditional manual deformation monitoring is increasingly automatized in order to be easily implemented and cost-efficient. The automatic selection of an adequate mathematic model in the data approximation, which includes stochastic and functional parts, is the basis of the automization of
the deformation monitoring procedure and will have influence on the reliability of the results. The thesis carries out some investigations in the above-mentioned background. In the initial study,
the risk of a commonly misspecified variance-covariance matrix (VCM), i.e. neglecting the mathematical correlations and assuming homoscedasticity, on the results of a congruency test, is high-lighted. The significant influence of a misspecified stochastic model on the deformation judgment motivates further investigations on a refined, heteroscedastic VCM based on a more detailed uncertainty budget of TLS measurements. The specified VCM can in generally be evaluated by means of two hypothesis testing procedures, i.e. a nested model misspecification test and a non-nested model selection test. In addition, the functional model also has strong influence on the deformation decision. Under this consideration, the more flexible B-spline models are applied in the
approximation and their performances are compared statistically with that of polynomial models in two case studies, where the superiority and limitation of them are exemplarily revealed. Besides the widely-used model selection procedures based on information criteria, we adopted two of the hypotheses test-based approaches, i.e. simulation-based Cox’s test and Vuong’s non-nested
test, to generally discriminate statistically between parametric models. The introduced automatic selection processes for the stochastic and functional models significantly improve the quality of the deformation monitoring process. It is therefore the basis for an interdisciplinary monitoring process.