Robust height estimation of moving objects from uncalibrated videos

Area of Science:

• Computer Vision
• Metrology
• Geometric Computer Vision

Background:

• Accurate 3D measurements from uncalibrated video are challenging.
• Existing methods often require manual calibration or supervision.
• Tracking moving objects on a ground plane offers geometric cues.

Purpose of the Study:

• To develop an automated video metrology system.
• To recover scene geometry (vanishing line, vertical point) from uncalibrated camera footage.
• To enable robust, flexible, and accurate height measurements.

Main Methods:

• Recovering vanishing line and vertical point using moving object trajectories on a ground plane.
• Employing a probabilistic model for trajectory grouping and vanishing point estimation.
• Applying single-view mensuration and fusing multiframe measurements with LMedS and RMSA techniques.

Main Results:

• Successful height measurements obtained from uncalibrated video.
• Demonstrated robustness and reduced human supervision.
• Uncertainty analysis confirms practical robustness of the auto-calibration method.

Conclusions:

• The proposed video metrology approach is effective and robust.
• It offers a flexible and less supervised alternative for 3D measurements.
• The method shows promise for various real-world applications.