- A 2005 paper reports on the benefits of using multi-spectral imagery combined with LiDAR to automatically identify buildings.
- The authors use the NDVI to determine surface roughness, which helped to distinguish trees from buildings.
- They were able to correctly identify 90% of the existing buildings.
I just came across a paper published in 2005 identifying the benefits of using multi-spectral imagery to improve the automatic detection of buildings. The authors note that the major challenge is in distinguishing between trees and buildings. One of the best cues for making that distinction is surface roughness. LiDAR data can provide this kind of information, but as the resolution decreases the problem becomes more difficult .
To overcome this the authors relied on the use of multispectral imagery and the NDVI – normalized difference vegetation index, a fairly common parameter used in the exploitation of multispectral imagery. They were able to detect 90% of the buildings in 2 different types of data sets, with the missed buildings tending to be smaller in footprint. Of course this kind of result may or may not be acceptable based on your project specification, but it certainly is worth being aware of this approach.
It would be interesting to know if the work has continued.
