Dynamic 3D Reconstruction of Vehicles for Safer Intersections

There have been concerted nationwide efforts to improve roadway safety. However, the fatality rate and number of fatalities have increased over the past couple of years from reckless behaviors such as speeding, alcohol impairment, and not wearing seat belts. Intersections are particularly dangerous sections of the roadway system due to being points of conflict between vehicles, pedestrians, and bicyclists. Multiple video cameras are becoming increasingly common at urban traffic intersections. The motion of the vehicles can be invaluable to traffic analysis, including vehicle type, speed, density, trajectory and frequency of events such as near-accidents. The authors present a comprehensive framework that fuses (a) incomplete and imprecise structured points (part detections) across multiple views with (b) precise but sparse single-view tracks of unstructured points, to reconstruct moving vehicles even in severe occluded scenarios. This framework consists of three main stages: (1) a novel object-centric (as opposed to featurecentric) RANSAC approach to provide a good initialization of the 3D geometry of the structured points of the vehicle, (2) a novel algorithm that fully exploits the complementary strength of the structured and unstructured points via rigidity constraints, and (3) closing-the-loop by reprojecting the reconstructed structured points to all views to retrain the part detectors.

  • Record URL:
  • Summary URL:
  • Supplemental Notes:
    • This document was sponsored by the U.S. Department of Transportation, University Transportation Centers Program.
  • Corporate Authors:

    Carnegie Mellon University

    Robotics Institute, 5000 Forbes Avenue
    Pittsburgh, PA  United States  15213-3890

    Technologies for Safe and Efficient Transportation University Transportation Center

    Carnegie Mellon University
    Pittsburgh, PA  United States  15213

    Office of the Assistant Secretary for Research and Technology

    University Transportation Centers Program
    Department of Transportation
    Washington, DC  United States  20590
  • Authors:
  • Publication Date: 2018

Language

  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; Photos; References;
  • Pagination: 9p

Subject/Index Terms

Filing Info

  • Accession Number: 01690846
  • Record Type: Publication
  • Report/Paper Numbers: Project ID: 30
  • Contract Numbers: DTRT12GUTG11
  • Files: UTC, TRIS, ATRI, USDOT
  • Created Date: Jan 10 2019 9:35AM