GIS and automatic object recognition

Document created by imostertman1975 on Jan 13, 2017
Version 1Show Document
  • View in full screen mode

Are documented assets really representing the actual world outside? To tackle this question we are developing an application for the Dutch rail infrastructure company that compares the geographic position of assets, such as railway signs, which are detected in video footage shot by inspection trains with the estimated position of identical assets as documented in topological CAD files.


Developed background processes combining smart Computer Vision algorithms with GPS data from inspection trains recognizes pre-defined assets with a position accuracy up to a couple of meters. With the help of background ETL processes to automatically process CAD files and data consumed from ArcGIS Server REST services, geographic positions of assets is estimated based on their described relative topological positions.


The end user application shows video footage and highlights inconsistencies between both worlds on the respective interactive (topological) map. Based upon the demonstrated information the user can analyze if eventual detected inconsistencies are valid.


This presentation will highlight the used techniques and challenges we encountered in developing this application.

5 people found this helpful