Camera pose estimation algorithm involving weighted measurement uncertainty of feature points based on rotation...

Area of Science:

• Computer Vision
• Robotics
• Photogrammetry

Background:

• The perspective-n-point (PnP) problem is fundamental in computer vision for determining camera pose.
• Existing PnP algorithms often require depth information or initial guesses, limiting their applicability.
• Accurate camera pose estimation is crucial for applications like augmented reality, autonomous navigation, and 3D reconstruction.

Purpose of the Study:

• To develop a novel camera pose estimation algorithm that addresses the limitations of existing PnP methods.
• To introduce a method that does not rely on depth information and can be solved directly without initial values.
• To improve the accuracy and robustness of visual measurement through weighted measurement uncertainty.

Main Methods:

• A camera pose estimation algorithm based on weighted measurement uncertainty using rotation parameters.
• Conversion of the objective function into a least-squares cost function involving three rotation parameters.
• Incorporation of a noise uncertainty model for enhanced pose estimation accuracy.

Main Results:

• The proposed method achieves direct calculation of camera pose without requiring initial values.
• Experimental results demonstrate high accuracy and robustness in camera pose estimation.
• Maximum estimation errors for rotation and translation were found to be better than 0.04° and 0.2% respectively in a 1.5m x 1.5m x 1.5m space.

Conclusions:

• The developed algorithm effectively solves the perspective-n-point problem in visual measurement.
• The method's independence from depth factors and its direct solvability offer significant advantages.
• The algorithm provides a robust and accurate solution for camera pose estimation, suitable for various computer vision applications.