An Accurate Linear Method for 3D Line Reconstruction for Binocular or Multiple View Stereo Vision

Area of Science:

• Computer Vision
• Geometric Reconstruction
• Photogrammetry

Background:

• 3D line reconstruction is crucial for stereo vision applications.
• Existing methods like Direction-then-Point (DtP) and Two Points 3D coordinates (TPS) have limitations when only one point correspondence is available.
• Accurate 3D line reconstruction is essential for tasks like robotics, augmented reality, and 3D modeling.

Purpose of the Study:

• To introduce a novel linear method, Point-then-Direction (PtD), for 3D line reconstruction using a single pair of corresponding points.
• To evaluate the accuracy of the proposed PtD method against established DtP and TPS methods.
• To demonstrate the suitability of PtD for multi-camera systems.

Main Methods:

• The Point-then-Direction (PtD) method utilizes the intersection of rays from corresponding image points to determine a 3D point's coordinate.
• The 3D point is then used as a positional constraint to calculate the line's direction by minimizing image angle residuals.
• The method was tested using simulations in binocular and multiple view stereo vision setups.

Main Results:

• PtD significantly improves the accuracy of both 3D line direction and position compared to the DtP method.
• PtD demonstrates superior accuracy in estimating the 3D line's direction compared to the TPS method.
• While PtD excels in direction accuracy, TPS offers higher precision for 3D line position.

Conclusions:

• The proposed Point-then-Direction (PtD) method offers a more accurate and linear approach to 3D line reconstruction with a single point correspondence.
• PtD provides a valuable alternative to existing methods, particularly when precise line direction is critical.
• The findings suggest PtD is a robust technique applicable to various stereo vision configurations, including multi-camera systems.