Realtime Obstacle Detection and Tracking Based on Constrained Delaunay Triangulation

Motivation and Approach

  • Vision-based obstacle detection and tracking from in-vehicle video.
  • Appearance-based detection: over-segmentation & grouping.
  • Realtime detection: fast segmentation based on Constrained Delaunay Triangulation.
  • Apply Support Vector Machine for hypotheses verification: over-segmentation & grouping approach enables the efficient use of texture features.
  • Robust dual-layer tracking with probabilistic reasoning: false detections are removed and tracking failure is compensated.

    • Detection Results (Good & Bad Results)





    Tracking Results

    Result for IEEE ITSC'06 (excerpt video clips) : vdetect.avi

    [For future comparative studies]
    Original video clip: samtrans.mpg
    Our result: samtrans-result.avi
    The final results are shown in cyan rectangles, backup candidates are in black, and the original frame-by-frame detection results are in white.

    Publications

  • Z. Kim, "Realtime Obstacle Detection and Tracking Based on Constrained Delaunay Triangulation", IEEE Intelligent Transportation Systems Conference, pp. 548-553, 2006. [PDF file]

    • [Home]