Geolocation of utility assets using omnidirectional ground-based photographic imagery

A process for using ground-based photographic imagery to detect and locate power distribution assets is presented. The primary feature of the system presented here is its very low cost compared to more traditional inspection methods, because the process takes place entirely in virtual space. Specifically, the system can locate assets with a precision comparable to typical GPS units used for similar purposes, and can readily identify utility assets, for example transformers, if appropriate training data are provided. Further human intervention would only be necessary in a small fraction of cases, where very high uncertainty is flagged by the system. The feasibility of the process is demonstrated here, and a path to full integration is presented.