top of page
image.png

Come and check our booth at GIS-T. The Geospatial Information Systems for Transportation Symposium (GIS-T) comprises the American Association of State Highway and Transportation Officials, state-level Departments of Transportation, the Federal Highway Administration, various transportation-related research councils, and transportation industry products and services providers. It brings together professionals from government and private industry interested in the use of GIS for transportation. Network and share experiences, knowledge, see state-of-art software, and learn more about this field. 

image.png

Presentation Title: Object tracking and geo-localization from street images. Daniel Wilson, Xiaohan Zhang, Kamiku Xue, Safwan Wshah Vermont Artificial Intelligent Laboratory Department of Computer Science University of Vermont.

​

VTrans Contacts: Rick Scott and Alex Geller.

​

Presentation slides: here  

image.png

2022 Symposium 

Brief Project Description: We have constructed an end-to-end system which receives road images captured from a vehicle as input and automatically constructs a GIS map by classifying signs and placing them at their geo-location. Our pipeline contains three core components. First, an object detector detects signs in an image, determines their class, and predicts their geo-location. Second, an object tracker reduces repeated detections of the same sign into a single prediction for each sign. Third, signs are displayed at their geo-location in an interactive visualization tool we have constructed.

Presentation slides: here  

image.png

2021 Symposium 

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.

Presentation slides: here  

image.png

2020 Symposium 

Brief Project Description: We have annotated the world's largest traffic sign recognition dataset containing 181 distinct classes of road signs and their GPS coordinates. From this data, we have built an automated system that identifies, classifies, and geolocalizes the signs from roadside images. This system will ultimately be capable of automatically constructing a GIS map displaying the type and GPS location of traffic signs using roadside images as input.

Presentation slides: here  

image.png

2019 Symposium 

Brief Project Description: Our research aims to develop an automated system that processes a stream of images and classifies the types of signs present in the image and determines the GPS location of that sign. Furthermore, our project introduces one of the few large-scale datasets to serve as a benchmark in the domain of Traffic Sign Recognition (TSR).

Presentation slides: here  

bottom of page