3D Pose Estimation by Directly Matching Polyhedral Models to Gray Value Gradients

  • Authors:
  • Henner Kollnig;Hans-Hellmut Nagel

  • Affiliations:
  • Institut fü/r Algorithmen und Kognitive Systeme, Fakultä/t fü/r Informatik der Universitä/t Karlsruhe (TH), Postfach 6980, D-76128 Karlsruhe, Germany;Institut fü/r Algorithmen und Kognitive Systeme, Fakultä/t fü/r Informatik der Universitä/t Karlsruhe (TH), Postfach 6980, D-76128 Karlsruhe, Germany/ Fraunhofer-Institut fü ...

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 1997

Quantified Score

Hi-index 0.00

Visualization

Abstract

This contribution addresses the problem of pose estimation andtracking of vehicles in image sequences from traffic scenes recordedby a stationary camera. In a new algorithm, the vehicle pose isestimated by directly matching polyhedral vehicle models to imagegradients without an edge segment extraction process. The newapproach is significantly more robust than approaches that rely onfeature extraction since the new approach exploits more informationfrom the image data. We successfully tracked vehicles that werepartially occluded by textured objects, e.g., foliage, where aprevious approach based on edge segment extraction failed. Moreover,the new pose estimation approach is also used to determine theorientation and position of the road relative to the camera bymatching an intersection model directly to image gradients. Resultsfrom various experiments with real world traffic scenes arepresented.