Shape-Based Alignment of the Scanned Objects Concerning Their Asymmetric Aspects

Area of Science:

• Computer Vision
• Robotics
• Industrial Automation

Background:

• Accurate 3D object recognition and alignment are crucial for automated industrial processes.
• Existing methods face challenges with symmetric or asymmetric object shapes and precise alignment.

Purpose of the Study:

• To develop an integrated method for processing laser depth maps for object recognition and alignment.
• To enhance the Triangle Area Representation (TAR) method for improved shape description and object boundary selection.
• To create a robust algorithm for calculating 3D rigid affine transforms for precise object alignment.

Main Methods:

• Utilizing Retinex filter for initial contour detection in depth maps.
• Employing an extended Triangle Area Representation (TAR) for accurate object boundary identification, including asymmetric features.
• Implementing a novel point-to-point pairing algorithm for precise 3D rigid affine transform calculation, surpassing the Kabsch algorithm's accuracy.

Main Results:

• Successfully implemented a pipeline for object recognition and alignment from single laser depth map examples.
• The extended TAR method effectively handles objects with both symmetric and asymmetric characteristics.
• The developed alignment algorithm achieves precise contour-to-contour matching for accurate 3D transformations.

Conclusions:

• The integrated method provides a robust solution for object recognition and alignment in industrial settings.
• The enhanced TAR method and novel alignment algorithm offer improved accuracy and versatility.
• This pipeline is well-suited for applications such as production inspection and quality control.