Multi-View Metal Parts Pose Estimation Based on a Single Camera.

Area of Science:

• Robotics
• Computer Vision
• Industrial Automation

Background:

• Accurate 6D pose estimation is crucial for industrial grasping of metal parts.
• Reflective properties of metal parts hinder complete point cloud acquisition.
• Existing methods often require depth information or struggle with shiny surfaces.

Purpose of the Study:

• To develop a method for recovering the 6-Degrees-of-Freedom (6D) pose of CAD-model-known metal parts using a single RGB camera.
• To overcome the challenges posed by reflective surfaces in obtaining accurate pose estimations.
• To enable pose estimation without requiring depth data.

Main Methods:

• Utilizes multiple views for pose estimation of metal parts.
• Employs ray casting to simulate additional views and determine the camera's next best viewpoint.
• Integrates camera transformations with multi-view pose data for final pose refinement.

Main Results:

• The proposed method effectively estimates the 6D pose of shiny metal parts.
• Demonstrates successful pose recovery using only RGB images.
• Achieves accurate results despite the reflective nature of the target objects.

Conclusions:

• The multi-view approach with ray casting is effective for 6D pose estimation of metal parts.
• The method provides a viable solution for industrial grasping applications requiring accurate pose data.
• RGB-only pose estimation is feasible for challenging reflective objects.