Adaptive Support Weight-Based Stereo Matching with Iterative Disparity Refinement

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Minimally invasive surgery (MIS) benefits from real-time 3D reconstruction for improved depth perception and navigation.
• Endoscopic imaging in MIS faces challenges like specular reflections, low-texture, and tissue deformation.

Purpose of the Study:

• To develop a novel, deterministic, and iterative stereo-matching method for accurate real-time 3D reconstruction in MIS.
• To address the specific constraints of endoscopic imaging for improved surgical visualization.

Main Methods:

• A deterministic and iterative stereo-matching algorithm utilizing adaptive support weights was developed.
• The method was implemented in CUDA and C++ for real-time performance.
• Evaluation was performed on the Stereo Correspondence and Reconstruction of Endoscopic Data (SCARED) dataset and a custom synthetic dataset.

Main Results:

• On SCARED datasets 8 and 9, the method achieved Mean Absolute Errors (MAE) of 3.79 mm and 3.61 mm, respectively, at 24.9 FPS.
• On synthetic data, the method demonstrated an MAE of 140.06 μm and Root Mean Square Error (RMSE) of 251.9 μm.
• Performance is comparable to or surpasses existing deterministic stereo-matching approaches in challenging endoscopic conditions.

Conclusions:

• The developed method provides accurate, deterministic, real-time depth estimation for MIS.
• It offers a foundation for single-shot 3D reconstruction, stereo frame stitching, and full-scene modeling.
• Potential applications include integration into surgical navigation, robotic assistance, and augmented reality systems.