Speedup 3-D Texture-Less Object Recognition Against Self-Occlusion for Intelligent Manufacturing

Area of Science:

• Robotics and Intelligent Systems
• Computer Vision
• 3-D Perception

Background:

• Real-time 3-D object detection and pose estimation are vital for intelligent manufacturing tasks like robot feeding and assembly.
• Detecting and estimating the pose of textureless metal parts in cluttered, occluded environments remains a significant challenge for current state-of-the-art methods.

Purpose of the Study:

• To propose an online 3-D object detection and pose estimation method specifically designed to overcome self-occlusion for textureless objects.
• To enhance robustness and efficiency in 3-D perception for industrial applications.

Main Methods:

• Utilizes raw 3-D point clouds with normal cues for local reference frame definition.
• Employs an autoencoder to learn compact 3-D features from local normal statistics.
• Introduces a novel basis buffer k-d tree method for efficient similarity search, maximizing GPU parallel processing.
• Generates hypothesis candidates via Hough voting, filters false hypotheses, and refines pose estimation using the iterative closest point strategy.

Main Results:

• The proposed method demonstrates effectiveness in detecting and estimating the pose of industrial objects with heavy self-occlusions.
• Experimental comparisons validate the superior effectiveness and efficiency of the developed method against existing state-of-the-art approaches.
• The basis buffer k-d tree method enhances practical GPU parallel processing capabilities.

Conclusions:

• The developed online 3-D object detection and pose estimation method provides a robust and efficient solution for handling textureless objects in cluttered industrial settings.
• The approach successfully addresses challenges posed by self-occlusion, paving the way for more advanced robotic manipulation and assembly.
• The method's ability to leverage raw 3-D point clouds and efficient feature learning contributes to its practical applicability in real-world intelligent manufacturing.