An Online 3D Modeling Method for Pose Measurement under Uncertain Dynamic Occlusion Based on Binocular Camera

Area of Science:

• Robotics
• Computer Vision
• Geometric Modeling

Background:

• 3D modeling is crucial for industrial pose measurements, but dynamic occlusions hinder online applications.
• Existing methods struggle with uncertain dynamic objects disrupting 3D model creation.

Purpose of the Study:

• To develop an online 3D modeling method robust to uncertain dynamic occlusions using a binocular camera.
• To enable accurate 3D reconstruction and pose measurement in dynamic industrial environments.

Main Methods:

• A novel dynamic object segmentation method using motion consistency constraints and pose clustering.
• An optimization technique incorporating local constraints in overlapping view regions and global loop closure for point cloud registration.
• Implementation and validation in a custom experimental workspace.

Main Results:

• Successful online 3D modeling under uncertain dynamic occlusion.
• Acquisition of complete and accurate 3D models.
• Effective pose measurement demonstrating the method's practical utility.

Conclusions:

• The proposed method effectively addresses dynamic occlusion challenges in online 3D modeling.
• It enables precise 3D reconstruction and reliable pose estimation for industrial applications.
• The approach offers a significant advancement for real-time geometric analysis in dynamic environments.