Enhanced distance perception in automated robotic endoscopy through scale-aware monocular depth estimation.

Area of Science:

• Robotics
• Computer Vision
• Surgical Technology

Background:

• Accurate distance perception is critical for safe and effective robotic surgery, especially in minimally invasive procedures.
• Current distance estimation methods often rely on expensive or ineffective sensors in endoscopic environments.
• Limited visual feedback in endoscopic surgery necessitates advanced perception capabilities for surgical robots.

Purpose of the Study:

• To develop a monocular image-based framework for accurate distance measurements in surgical robotics.
• To enhance the autonomous navigation capabilities of surgical robots in complex anatomical structures.
• To overcome the limitations of existing distance estimation methods in endoscopic settings.

Main Methods:

• A monocular endoscopic image-based pipeline for scale-aware depth estimation.
• Utilizing surgical instruments with known cylindrical geometry for absolute scale recovery.
• An automatic module for continuous tracking of tissues and instruments for precise distance calculations.

Main Results:

• Reduced mean absolute error from 8.56 mm to 6.35 mm compared to baselines.
• Improved depth estimation accuracy from 76.9% to 88.0% with near-perfect absolute scale.
• Achieved average distance errors between 4.7 mm and 6.4 mm in complex endoscopic scenarios.

Conclusions:

• A novel image-based distance perception framework enhances surgical robot automation.
• The framework enables efficient object distance measurements within endoscopic views.
• Demonstrated potential to benefit robotic surgery tasks, including automatic navigation.