Recognition and Grasping of Disorderly Stacked Wood Planks Using a Local Image Patch and Point Pair Feature Method.

Area of Science:

• Robotics
• Computer Vision
• Machine Learning

Background:

• Robot recognition and grasping of irregularly stacked objects, like wooden planks, is a challenging task.
• Existing methods often struggle with variations in object pose and environmental clutter.

Purpose of the Study:

• To develop an effective recognition and positioning method for robots to grasp disorderly stacked wooden planks.
• To improve the accuracy and success rate of robotic grasping in complex industrial or construction environments.

Main Methods:

• Utilized a convolutional autoencoder to extract robust local texture feature descriptors from image patches.
• Combined local image features with point pair geometric features to create a comprehensive feature description code for wooden planks.
• Employed point pair feature matching, pose voting, and clustering for precise pose determination of the target plank.

Main Results:

• Achieved a high plank recognition rate of 95.3%.
• Demonstrated a successful grasping rate of 93.8% in experiments.
• The proposed method showed significant advantages over traditional point pair feature (PPF) methods.

Conclusions:

• The proposed method is effective for recognizing and determining the pose of wooden planks in cluttered, stacked environments.
• This technique offers a robust solution for robotic grasping applications involving challenging object arrangements.
• The high success rates indicate its practical applicability in real-world scenarios.