Skip to main content
ICTD 2024 Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
ViewAttendees 1

Abstract for presentation (Poster or Podium)

AI in Transportation

An Innovative System for Automatic Traffic Sign Detection and Recognition with Monocular Camera

Monday, June 17, 2024
1:30 PM - 2:45 PM ET
Room: Galleria
Abstract: Traffic signs inventory is essential for asset management as it provides information about traffic sign, such as the installation date, type, and condition. Previously, many road agencies manually collected traffic signs data, which is a time-consuming process and has low accuracy. A computer vision algorithm has been applied to facilitate traffic sign detection and recognition tasks and solve labor-intensive issues. As public traffic signs datasets were collected in different regions, the existing models cannot be directly used to detect and recognize US traffic signs. Therefore, this study collected an additional 5,000 traffic signs in Washington area from Google Map API and self-installed dash cameras. Then, these data were manually labeled into 43 classes for training. To develop a traffic sign detection and recognition model (TSDR), the Faster R-CNN Inception V2 is selected as a base model for detection with an accuracy rate of 98.34%. Existing datasets and collected data were used to develop a customized traffic sign recognition model, which yielded an accuracy of 97.1%. In addition, an automated pipeline for traffic signs captures, detects, classifies, and stores is developed. By embedding the TSDR model in edge devices, this system allows the ability to uphold privacy standards. Processed data, consisting of an image and type of traffic sign, is transmitted to the server automatically. A sample traffic signs data inventory in Washington state was created. This database is a valuable resource for developing and testing various machine-learning models. Additionally, it can be used for asset management

Learning Objectives:

Attendees can expect to learn the following from this session:
  • Improving existing traffic sign detection and recognition models with our self-collecting data, which has better accuracy for US traffic signs.
  • Develop edge devices to provide cost-efficient and preserve privacy in data collection.
  • Develop the prototype of the Washington State Department of Transportation traffic signs data inventory system.