Automating detailed urban data extraction from high resolution aerial imagery

Ordnance Survey are partners in the CityVerve project in Manchester. This project is building a city Internet of Things demonstrator, led by Manchester City Council and Cisco. OS is supplying geospatial data to CityVerve. We are interested in examining the framework of data needed for the detailed monitoring of the city environment allowed by an IoT system and to that end have acquired 4cm ground-resolution aerial imagery. In association with this our remote sensing and surveying teams have acquired especially detailed topographic mapping of the study area (see the attached imagery), including for example nearly 8,000 utility covers, 3,000 posts and over 1,000 cycle racks. However this effort is not easily scaled to the rest of the country. Given this available data, what can be done to scale up? Can machine learning techniques, using the current data to train the system, be used? Is there a role for crowd-sourcing and citizen science? Or both techniques together?