Verification of 3D Building Models using Mutual Information in Airborne Oblique Images

  • Mr Adam Nyaruhuma, Faculty of Geoinformation Science and Earth Observation (ITC), University of Twente, The Netherlands
  • Markus Gerke, University of Twente, Faculty ITC, Department of Earth Observation Science, The Netherlands
  • George Vosselman, Faculty of Geoinformation Science and Earth Observation (ITC), University of Twente, The Netherlands

This paper describes a method for automatic verification of 3D building models using airborne oblique images. The problem being tackled is identifying buildings that are demolished or changed since the models were constructed or identifying wrong models using the images. The models verified are of detailed wire-frame type where corresponding face edges on the models and in images can be recognized. The method utilizes multiple-perspective overlapping images. The verification approach is based on Information Theory. Corresponding variables between building models and oblique images are used for deriving Mutual Information. The Mutual Information is obtained for individual edges, or faces of a building model, and combined for all perspective images available for the whole building.
The wire-frame model edges are projected to images and verified using low level image features - the image pixel gradient directions. The gradient directions are utilized as evidence of model edges. An important step in the proposed method is visibility analysis. A building model part is only checked against images in which it may be visible.
The method has been tested using Pictometry images that are available for most cities of Europe and may be publicly viewed in the so called Birds Eye view of the Microsoft Bing Maps. Images of a part of the city of Enschede, the Netherlands were acquired, oriented and experiments conducted for 105 existing and 117 demolished buildings. Because of a limited number of demolished buildings in the experimented area, most of these are simulated.
Results are: all buildings that are categorized as existing actually exist, i.e. no demolished building is wrongly identified as existing and only six existing buildings are wrongly categorized as demolished. The six models are also either wrong or shifted such that their edges do not coincide with edges of faces of buildings in the images.
These results indicate that overlapping oblique images are useful for identifying buildings that were or were not changed since models were acquired. The changes may have occurred due to urban redevelopment or during disasters such as tsunami or earthquakes.
The results reported here were obtained only considering roofs. We intend to extend the method in order to take advantage of an important characteristic of oblique images: the presence of both top (roof) and side views (walls) of buildings. Besides overall building verification, results obtained per individual model edges or faces can be used for checking and improving the 3D building models. For example, if a building exists and most faces are correct, the overall verification results for the model will indicate that the building exists although individual parts of the model may have been wrongly captured and their verification results are contrary to the overall. Although the wrong parts may be manually updated, automatic refinement of 3D building models using oblique images is an interesting future work.

Building verification results Results existing Results Demolished
Actually Existing 107 6
Actually Demolished 0 109
Total 107 115
Correctness_existing TP/(TP+FN) 1.0
Correctness_Demolished TN/(TN+FP) 0.95