This AI Pack can be purchased as an add-on to the Building Footprints pack. It adds metadata to the building footprints for:
- The "Peak Height" of the building
- The number of storeys of the building
AI Feature API
(Gen 1 & 2 Data)
Building Characteristics Spec
Metadata attached to Building Feature Class
- Storeys (1, 2, 3+) with confidences
- Peak Height
| No applicable AI Layers|| Not available in Gen 1 / 2 Data|
*All data are based on WGS84 (EPSG:4326)
3D Building Characteristics
The 3D building characteristics are derived using Nearmap’s 3D data and supported in surveys with 3D capabilities.
From Gen 3 to Gen 5
gen5-tranquil_sea-1.0, they were calculated by finding the closest date of mesh to the "survey date" being executed. As not all our surveys captured 3D imagery, the 3D date was not always identical to the survey date the images/AI were getting pegged against. Where a 3D-enabled survey was flown, it was usually the 3D produced on the same survey.
From the Gen 5
gen5-urban_atlas release, the above approach has been adjusted to process 3D attributes only on surveys with a precise match between the 3D model and Vertical imagery date.
Some Gen 3 building footprints in Phoenix, AZ overlaid on our webtile imagery, and extruded using the Peak Height in a GIS application.
This is the height from a ground point nearby the building to the 95th percentile of the roof height. This is essentially the ground to the peak of the roof, with some robustness to protrusions or noise that may be attached to the roof.
A ‘storey’ is a building level that is above ground and enclosed by walls. Where the top storey is a partial storey, the algorithm should detect it as two storeys. A slightly elevated building with some crawlspace under the building still only counts as a single storey building. A storey can include rooms, garages: any space that is part of the dwelling.
The storey count is a categorical multi-class attribute, with values and confidences returned for 1, 2 or 3+ storeys. This is not simply an approximation based on Peak Height - we apply machine learning algorithms to our 3D models, in order to accurately assess the number of storeys. The confidences sum to 100% and usually the value with the highest confidence is correct. e.g. if the results are 1 - 90%, 2 - 5%, 3+ - 5%, there is a 90% chance it is a single storey building. In situations where the confidence is split fairly evenly between two storeys (such as 1 - 39%, 2 - 41%, 3+ - 20%), it may indicate that the building has a half storey (in this case it may be a single storey base, with an attic, or partial second storey that does not cover the whole building).
This graph shows how the height in metres is related to the number of storeys, but is not simply a cutoff. We apply machine learning algorithms that consider the geometry of the building as a whole, not just the peak height. Data taken from an area within Phoenix, Arizona.
| These buildings are technically single storey, as the space below them is not habitable. |
| These buildings are two storey, because they contain components which have two enclosed, habitable floors. |
A difficult example - 2 storeys on the right-hand side, and one storey on the left. The algorithm should class this as a 2 storey building.