Year: 2014
Type: other
Abstract: Large scale collection of spatial data via mobile sensors allows us to better model the complex phenomena in our physical surroundings. This capability puts a strain on existing data storage infrastructures; and although the resulting models are much richer and closer to reality, these improvements come with the cost of added complexity. This thesis addresses needs to improve the accuracy, effectiveness, and efficiency of important spatial queries on new, information-rich, and complex models. Specifically, it addresses the need for speeding up existing spatial queries on 3D surfaces, for efficient spatial joins between 3D and 2D datasets, and for maintaining efficiently the up-to-date fastest paths for a large number of users in a dynamic spatial network. First, we propose lower bounds for the shortest surface distance, which in turn improves the performance of many important queries to which shortest surface distance computation is inherent. We propose two approaches. In the first, the lower bound is a ratio of the shortest network distance computed on the graph representation of the 3D triangulation. The second approach proposes further segmentation of existing triangles/faces so that a new surface face-crossing path is used to compute an even tighter lower bound. Experiments assist us in better understanding the relationship between lower bound tightness and query performance. Next, we present a framework that generates a 3D spatial network by conducting a spatial join between a massive 3D point-cloud (terrabytes) and a 2D spatial network (gigabytes). In a filtering phase, the 3D points close to the 2D road network are identified. Then, the subsequent lifting phase triangulates these to obtain elevation values of points in the 2D road network. We study two filtering approaches. The first scans the 3D point-cloud in a single pass and constructs the 3D road network on the fly, while the second approach reads the 3D points as disk blocks in accordance to a locality-preserving space-filling curve. Results from empirical studies with real- world data offer insight into the effectiveness and efficiency of the two approaches. Finally, we propose a scalable distributed system that maintains a large number of continuous fastest-path queries on a large spatial network, providing users with guaranteed approximate fastest paths to their destinations, while facing heavy real- time traffic updates. The solution assumes that the dynamic real-valued edge weights are bounded in a manner consistent with real-world data, which enables a theoretical guarantee on the travel time. Comprehensive experiments suggest that the proposed methods are accurate, efficient, robust, and scalable.
Author: Kaul, Manohar
Institution: Aarhus University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4412320672
