As featured in the February 22, 2022 Research and Innovation Newsletter: An automated tool that geo-locates traffic signs using roadside images will be highly beneficial for inventory assessments and maintenance plans.
Brief Project Description: We present a two-stage framework that detects and geolocalizes traffic signs from street videos. Our system uses a modified version of RetinaNet (GPS-RetinaNet), which predicts a positional offset for each sign relative to the camera, in addition to performing the standard classification and bounding box regression. Candidate sign detections from GPS-RetinaNet are condensed into geolocalized signs by our custom tracker, which consists of a learned metric network and a variant of the Hungarian Algorithm. To build this model, we have annotated a large dataset containing a broad distribution of traffic signs in a diverse set of environments.
Poster: 
Fact Sheet Video Other Q&A
Presenter: Daniel Wilson, UVM
2019 Symposium Fact Sheet Poster Video
Researchers' Note (2/7/2022): Currently, we are working on replacing RetinaNet with the Swin-Transformer architecture to enhance sign detection performance, and plan to implement semi-supervised learning techniques to take advantage of the large library of unlabeled images we have access to.