Learning to Recognize Distance to Stop Signs Using the Virtual World of Grand Theft Auto 5

This paper examines the use of a convolutional neural network and a virtual environment to detect stop signs and estimate distances to them based on individual images. To train the network, the authors develop a method to automatically collect labeled data from Grand Theft Auto 5, a video game. Using this method, they collect a dataset of 1.4 million images with and without stop signs across different environments, weather conditions, and times of day. Convolutional neural network trained and tested on this data can detect 95.5% of the stops signs within 20 meters of the vehicle with a false positive rate of 5.6% and an average error in distance of 1.2m to 2.4m. The authors also discovered that the performance their approach is limited in distance to about 20m. The applicability of these results to real world driving are promising and must be studied further.

• Supplemental Notes:
  • This paper was sponsored by TRB committee AHB30 Standing Committee on Vehicle-Highway Automation.
• Corporate Authors:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001
• Authors:
  • Filipowicz, Artur
  • Liu, Jeremiah
  • Kornhauser, Alain
• Conference:
  • Transportation Research Board 96th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2017-1-8 to 2017-1-12
• Date: 2017

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; Photos; References; Tables;
• Pagination: 16p
• Monograph Title: TRB 96th Annual Meeting Compendium of Papers

Subject/Index Terms

• TRT Terms: Autonomous vehicle guidance; Distance perception; Neural networks; Stop signs; Virtual reality
• Uncontrolled Terms: Video games
• Subject Areas: Highways; Operations and Traffic Management;

Filing Info

• Accession Number: 01631454
• Record Type: Publication
• Report/Paper Numbers: 17-05456
• Files: TRIS, TRB, ATRI
• Created Date: Mar 30 2017 3:11PM