Efficient global stereo-matching method of general images under a long baseline based on baseline estimation

Area of Science:

• Computer Vision
• 3D Measurement
• Photogrammetry

Background:

• Stereo matching for large 3D objects requires long baselines to minimize splicing errors, but this reduces efficiency.
• Short baselines improve matching efficiency by avoiding large disparity levels, applicable to both translation and rotation configurations.

Purpose of the Study:

• To develop an efficient stereo matching method for large 3D object measurement using short baseline estimation.
• To address the trade-off between accuracy and efficiency in stereo matching for large-scale applications.

Main Methods:

• A rectification method for three images was developed to create coplanar image planes, eliminating vertical disparity.
• The rectified images form stereo pairs with both short and long baselines.
• The relationship between disparities on different baseline stereo pairs was deduced, accounting for quantization errors.

Main Results:

• The disparity search range for long baseline stereo pairs was significantly reduced from hundreds to a few pixels.
• This reduction is based on predicted disparities from the short baseline pair.
• The method demonstrated improved calculation efficiency and accuracy in experimental results.

Conclusions:

• The proposed method enhances stereo matching efficiency for 3D measurement of large objects.
• It effectively reduces the disparity search range by leveraging short baseline estimation.
• The approach is suitable for practical applications requiring high efficiency and accuracy.