Steel Roll Eye Pose Detection Based on Binocular Vision and Mask R-CNN

Area of Science:

• Robotics and Automation
• Computer Vision
• Industrial Engineering

Background:

• Steel coil packaging requires precise inner corner guard installation.
• Manual installation is labor-intensive and prone to errors.
• Automation is needed to improve efficiency and reduce costs.

Purpose of the Study:

• To develop an automated binocular vision method for steel coil roll eye position detection.
• To enable automatic docking and installation of inner corner guards.
• To improve the accuracy and efficiency of steel coil packaging processes.

Main Methods:

• Utilized a deep learning framework with the Mask R-CNN algorithm for target region identification and mask image generation.
• Employed RGB vector space image segmentation and Sobel edge detection for feature extraction.
• Applied the least-squares method to fit parameters and determine deflection angle and center point coordinates.

Main Results:

• Achieved maximum deviations of 0.49 and 0.47 pixels for center point coordinates in the u and v directions.
• Demonstrated a maximum deflection angle error of 0.45° in ellipse parameter extraction.
• Exhibited maximum deviations of 2.17° (pitch), 2.24° (deflection), and 3.53-4.67 mm (centroid) in eye position detection, meeting installation requirements.

Conclusions:

• The proposed binocular vision method offers high accuracy and operability for steel coil roll eye position detection.
• This approach significantly enhances work efficiency and reduces labor costs in steel coil packaging.
• The method ensures adequate detection accuracy, fulfilling practical installation needs.