RBE/CS549 Spring 2024 - Class 10: Epipolar Geometry and Two-View Stereo

Overview

This class session focuses on the concepts of epipolar geometry and two-view stereo, which are fundamental in the field of computer vision and robotics.

Key Concepts

Epipolar Geometry

• Definition: Epipolar geometry is the geometry of stereo vision. When two cameras view a 3D scene from different positions, the geometry of the scene can be described by epipolar geometry.
• Components:
  • Epipolar Plane: A plane that contains the baseline (line connecting camera centers) and the 3D point being viewed.
  • Epipolar Line: The intersection of the epipolar plane with the image plane.
  • Epipole: The point of intersection of the line joining the camera centers with the image plane.

Fundamental Matrix

• Functionality: Encodes the intrinsic projective geometry between two views.
• Properties:
  • Relates corresponding points in stereo images.
  • Allows the computation of epipolar lines.

Essential Matrix

• Relation to Fundamental Matrix: Related but specific to calibrated cameras (removes intrinsic parameters).
• Use: Used to recover the relative rotation and translation between the cameras.

Stereo Vision

• Purpose: To perceive depth from two different viewpoints by matching corresponding points in the images.
• Challenges:
  • Finding corresponding points accurately.
  • Handling occlusions where points visible in one view are not visible in the other.

Two-View Stereo

• Process:
  • Image Rectification: Aligning images to simplify the matching process.
  • Depth Map Extraction: Using disparity between images to calculate depth.
• Applications: Critical in robotics, autonomous vehicles, and 3D reconstruction.

Practical Considerations

• Calibration: Ensuring camera parameters are correctly set for accurate stereo vision.
• Algorithmic Choices: Selection based on application requirements, such as speed and accuracy.

Conclusion

Understanding epipolar geometry and two-view stereo is essential for applications requiring depth perception and 3D reconstruction. Mastery of these concepts enables advancements in robotics and computer vision technologies.