Traffic Sign Recognition

state a fact about the project

TxDOT San Angelo District staff have been advancing the development and use of a nighttime sign inspection software tool.  The software runs while a multi-person van crew visually evaluates signs at night.  The crew updates a database with signs needing attention in real time (where communication is available).  An improvement to the tool would be an inventory of signs with locations and an image of the sign.  Incorporating the database into the tool would help the crew prepare for upcoming evaluation events in their limited-sight environment. 

Navigation Sign Identified at 96% Confidence

The project is to develop a method to economically create a sign database (inventory) for use with the inspection tool and other applications.  TxDOT has video coverage of district roadways via PathWeb, a product from Pathway Services Inc.  The concept is to use PathWeb content as input and artificial intelligence/machine learning techniques to identify traffic signs, capture an image of the sign, and capture the location coordinates.  These data elements are the prime material for an inventory.

The AI will be initially trained to identify a subset of traditional traffic signs.  Texas-based signage and markings will be extracted from Pathweb for this purpose with the research team labeling portions of the source images that contain the defined signs.

The following are examples of Pathweb video images analyzed by the AI model with sign identifications in the markup.  The decimal number is a confidence value.

Speed Limit Signed Identified with 98% Confidence
Stop Sign Identification at a Rural Intersection with 94% Confidence
Google Earth GIS Tool showing N. Mesa St in El Paso and a Geolocated Sign

Circling back to the original mission of supporting the nighttime sign inspection team, the AI system was given video camera input representing a typical trip.  The following short videos shows some research results.