Where Does It Belong? Autonomous Object Mapping in Open-World Settings.

Area of Science:

• Robotics
• Computer Vision
• 3D Reconstruction

Background:

• Detecting object changes (moved, removed, new) is vital for indoor robotics applications like tidying and patrolling.
• Open-world scenarios present challenges due to the unpredictable appearance of novel objects.

Purpose of the Study:

• To develop a robust method for detecting object changes from partial 3D reconstructions.
• To categorize objects into static, moved, removed, and novel states, even with incomplete data and clutter.

Main Methods:

• Identify planes in 3D reconstructions and consider clusters as objects.
• Compute point-pair-features for robust object matching and categorization.
• Utilize change detection knowledge without requiring specific object recognition.

Main Results:

• The proposed method effectively detects and categorizes object changes in challenging real-world scenarios.
• Outperforms a baseline learning-based object detection method in most test cases, even with limited training data.
• Demonstrates effectiveness in real robot experiments.

Conclusions:

• The developed approach provides a robust solution for object change detection in dynamic indoor environments.
• Enables robots to adapt to novel objects and environmental changes without prior specific object knowledge.
• The novel ObChange dataset facilitates quantitative evaluation of change detection methods.