DETECTING TOPOGRAPHIC CHANGE USING DIGITAL PHOTOGRAPHY AND DIGITAL SURFACE MODELS

摘要

The detection of changes to buildings is essential for the Ordnance Survey in order to maintain its national mapping database. This research contributes towards the efforts of the Research team at Ordnance Survey to create an automatic/semi-automatic tool for change detection. This paper outlines the use of DMC (Digital Mapping Camera) imagery to detect changes to buildings. At present the research has focussed on classifying the images and comparing the classification results to the map data to identify discrepancies. Three different classification methods have been tested, the one-class classifier Support Vector Data Description (SVDD), CART decision tree and Object-Based Classification (using Definiens Professional). These methods were initially tested using only the 4-band image, but results showed that incorporating a Digital Surface Model (DSM) derived automatically from the stereo imagery greatly improved classification accuracy and therefore the potential to use these classifications for change detection. For example, by using an object-based classification as an input to a change detection procedure we were able to detect 49 of the 51 significant changes to buildings on an image, though an additional 130 objects were flagged as changes but turned out to be false alarms. Work is ongoing to further enhance the filtering of predicted changes to reduce these false alarms. We are also investigating the use of images and DSMs from two different dates as an alternative approach to change detection. Research is now concentrating on evaluating which of these techniques would be most effective in a production environment.