Prism-empowered virtual multiview stereo for three-dimensional vision reconstruction.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Optical Engineering

Background:

• Traditional 3D vision reconstruction faces challenges balancing structural compactness, high resolution, and viewpoint flexibility.
• Achieving accurate 3D scene understanding often requires complex multi-camera setups or compromises in reconstruction quality.

Purpose of the Study:

• To develop a prism-empowered virtual multiview stereo vision architecture for 3D reconstruction using a single stationary camera.
• To present a variable-boresight perspective projection model for multiview image acquisition.
• To propose a model-driven performance analysis and error suppression strategy for improved reconstruction accuracy.

Main Methods:

• A rotational wedge prism combined with a camera to create a variable-boresight projection model.
• Model-driven analysis considering systematic and random errors for error suppression via cross-view constraints.
• Automatic calibration of the prism-driven dynamic virtual camera using motion estimation and intrinsic optimization for flexible multiview stereo matching.

Main Results:

• Demonstrated a prism-empowered architecture enabling 3D reconstruction with a single camera.
• Developed a reliable and robust multiview epipolar geometry through automatic calibration.
• Significantly enhanced the accuracy of stereo matching and the efficiency of object reconstruction.

Conclusions:

• The proposed prism-based virtual multiview stereo vision system offers a compact and flexible solution for 3D reconstruction.
• Automatic calibration and error suppression strategies lead to improved accuracy and efficiency in 3D scene understanding.
• This approach advances single-camera 3D reconstruction capabilities for various applications.