Coping with distance and location dependencies in spatial, temporal and uncertain data

The amount of collected data nowadays grows in an exponential manner due to a rapid development of capturing (such as photo cameras, environmental sensors or smart phones) and storage devices. Efficiently querying the resulting datasets is a crucial operation in order to retrieve meaningful information and to support time consuming data mining tasks. From a database perspective the efficient processing of queries is further facilitated by two circumstances. First the data becomes more and more complex and is not only given by single values but can amongst others be multidimensional, time-dependent or uncertain. Second the issued queries on these data types themselves need to be more sophisticated since users want to tap the full potential of the available information. This thesis focuses on three complex types of data, namely spatial, uncertain spatial and uncertain spatiotemporal data. For each of these data types certain dependencies are identified which can be utilized for more efficient or more accurate query processing. The first part of this thesis builds the basis for all following parts. It will give an introduction to the basic concepts of similarity search in databases. Therefore we will review the similarity model based on feature vectors and discuss several similarity query types on multidimensional data. Furthermore we will discuss basic concepts for efficient similarity query evaluation. During similarity query processing an important step is to filter out true drops as fast as possible. In the second part of this work we introduce the concept of spatial domination which can be utilized for filtering during several similarity queries. State-ofthe-art techniques for detecting spatial domination are either not applicable in all scenarios or do not accurately detect it. The reason for the latter problem are so called distance dependencies which are ignored in current methods. Thus this part successively gathers new techniques which overcome the present limitations in different scenarios. At the end of this part a technique is presented which is optimal w.r.t. runtime and accuracy. In the third part of this thesis the developed techniques from the second part are evaluated in the scope of uncertain spatial data. Again the concept of spatial domination based on distance dependencies can help to filter out objects during query processing. However the nature of uncertain objects requires the techniques for spatial data to be adapted. Thus a method for performing probabilistic domination is introduced. The main issue here is how to set off the single results against each other. For this purpose a technique called “uncertain generating function” is developed and evaluated. In the last part the focus is changed to uncertain spatio-temporal (UST) data. An