UC San Diego

This thesis investigates the feasibility of automated localization of light sources in the built environment.The aim is to enable at-scale geolocation of light sources with low-cost systems. The primary motivation is in support of a larger objective to later extract key parameters of electric grid performance from the discovered lights for wide-area, fine-grained grid health monitoring. Our findings suggest that it is possible to locate light sources in images and estimate the location reliably in the relative scenario, i.e., when the camera position is known a priori. However, we find that out-of-the-box supporting sensors---GPS for position and IMU for orientation---provide data with insufficient precision and accuracy for the final mapping task. We finish, then, with a discussion of how further innovations in image processing can help overcome these limitations to accurately resolve the real world position of lighting in the built environment.