Discovery and geolocation of assets from optical street level imagery
Infrastructure and utility companies currently use manual campaigns to populate and maintain asset databases. Assets such as road signage and telegraph/electricity network poles are usually distributed over large geographic areas making the process of data collection and monitoring labor-intensive and time-consuming. The challenge was to develop an automatic solution for the discovery and geolocating (mapping) of stationary street-side objects from optical and/or multisensor imagery.
AIMapIT is an innovative solution for the detection and GPS mapping of objects from street-level optical imagery. The system can be trained to detect any street furniture, vegetation, facade elements, and landmarks from optical imagery.
Inputs: Street view imagery openly available from Google, Bing, and Mapillary.
Technology: State-of-the-art deep learning elements for image analysis and an innovative geolocation module.
Outputs: Object detection and geotagging.
Results: Accurate and reliable mapping system: 95+% object detection precision and recall; 2m position (GPS) accuracy.
Prototype for pole detection developed for EIR
Detection accuracy 92+%.
Position accuracy within 2m.
Low compute complexity 200km of roads= 1h desktop/1-GPU
Awards: Shortlisted for two AI awards in 2017-2018
Reliable low-cost asset detection and monitoring.
Replacement of a costly and time-consuming manual process.
Enables recycling of existing image data.
Autonomous navigation – Drones and self-drive vehicles.
Agriculture and environment – hedge grows, coastal erosion, Japanese knotweed.
Street furniture inventory, monitoring, and maintenance.
Logistics – route planning.