Image-Based Change Detection Using An Integrated Spatiotemporal Gazetteer

This thesis addresses image-based change detection. Motivation was provided by the lack of algorithms that incorporate in their solution diverse types of pre-existing and complementary information and have the ability to interact with a spatiotemporal environment. The main differentiation with our approach is that we develop our algorithm within an integrated spatiotemporal environment and we make use of all change evidence that might exist within that environment. In addition, a change resolution model is developed that will distinguish meaningful changes based on user requirements. A model for change is proposed that establishes a general framework for the incorporation of image analysis techniques. This model is based on the SpatioTemporal Gazetteer. A theoretical analysis of change modeling resulted in the Differential Spatiotemporal Gazetteer. This modified version of the original gazetteer is composed conceptually of a Geographic Identity Information Level, a Change Indexing Information?level and a collection of child sources at the Child Information Level. Dependencies between these components are presented. In addition to that, two new components are introduced, the Change Detection Tools and the Knowledge Base expanding the gazetteer to a broader spatiotemporal environment. Information flow within the environment is described and component interaction is presented, focusing on the change detection aspects. In spatiotemporal applications, meaningful changes vary according to object type, level of detail, and nature of application. To compensate for this, we introduce two user-defined functions, the Minimum Spatial Element and the Minimum Temporal Element. These functions act as thresholds in the spatial and temporal domain, respectively, and allow the user to establish a resolutional framework in the spatiotemporal domain. With the proposed spatiotemporal model, we improve the change detection process, by providing validity of the datasets used, accuracy during the detection process, and a framework for storing the obtained results. The digital image analysis method that was developed automatically identifies object outline changes in sequences of digital images. This change detection technique is based on least squares template matching (LSM). We extend LSM to function in a differential mode. In doing so, we integrate object extraction from digital imagery with change detection in a single process. By using image orientation parameters and positional data we can reduce the problem of 3-D object monitoring to an image-space 2-D matching problem. In this hybrid approach, area- and feature-based matching are combined in one step, since raster and vector datasets are integrated to enhance our solution. Analysis of the edge geometry within a template, before matching takes place, improves the accuracy and reliability of the presented technique based on response time and obtained results. As a post-process, actual change is distinguished from different representations of the same object due to sensor inaccuracies, through fitting models of known systematic photogrammetric errors in the exterior and interior orientation process. The obtained results update the corresponding components of our spatiotemporal model through the detected change information. In doing so we establish a spatiotemporal model that makes use of the change detection results