Is it a shop or a house?

Aerial imagery most often looks down at objects on the ground whether its acquired from a UAV, manned aircraft or satellite. While there is no doubting the utility of PLACE Aerial imagery as seen from the results of models run in Malawi, Turks and Caicos and Anguilla, by looking down on something you can't necessarily tell what it is. A roof is a roof whether it is a commercial building or not. While scale of a building will give you a clue as to its use, for example a hospital, in a dense urban fabric determining whether a building is a stall or a home can be more difficult.

That's where PLACE Ground comes in. PLACE also acquires high resolution 360-degree terrestrial imagery. This combined with aerial imagery enriches data collection by giving a horizontal perspective so planners see what a building is being used for. Not only can this help with population estimates by excluding commercial buildings from head counts, it can also help with permitting and zoning compliance.

So how do you combine aerial and terrestrial imagery? We use Esri's  Orientated Image Catalog  (OIC). As both our aerial and street image data is geo-tagged (i.e. is positionally accurate), the OIC allows us to pair both datasets to see objects from the air and the ground. Not only that, using an OIC you can map, measure and describe features in street imagery and have that information appear on a map so you can quickly build a point of interest (POI) dataset for planning purposes.

The video below shows an OIC for Abidjan, Ivory Coast. This combines PLACE Aerial and Ground data collected for the city:

We use machine learning (ML) on our terrestrial imagery too, after all an image is an image no matter if its taken from the air or the ground!

One of the most important things PLACE does is to de-identify street imagery that means blurring faces and number plates to remove personally identifiable information (PII). We're using  Yolo  (You only look once) in combination with labeling and blurring tools to do  this , but there are other things that can be done with ML and street imagery like detecting  traffic lights .