Reconstruction-based 6D pose estimation for robotic assembly.

Area of Science:

• Robotics
• Computer Vision
• Industrial Automation

Background:

• Accurate pose estimation is crucial for robotic tasks like bin picking and assembly.
• Industrial environments present challenges due to diverse object shapes and complex settings.

Purpose of the Study:

• To develop a robust and accurate method for estimating the poses of narrow and elongated industrial objects.
• To guide robotic assembly processes using low-cost RGB-D cameras.

Main Methods:

• A three-step approach: preprocessing reconstruction, pose initialization using geometric features, and contour-aided tracking.
• Real-time dense reconstruction to create smooth depth images.
• Extraction of 3D planar features for initial pose estimation.
• Combining dense surface and sparse contour correspondences for robust pose refinement.

Main Results:

• The proposed method demonstrates feasibility in estimating object poses.
• The approach shows satisfactory tolerance to initial pose guesses.
• Successful application for guiding robotic assembly of industrial parts.

Conclusions:

• The developed method offers an effective solution for pose estimation of challenging industrial objects.
• Utilizing geometric features and contour cues enhances tracking robustness.
• Low-cost RGB-D cameras can be effectively employed for precise robotic assembly guidance.