Registration Combining Wide and Narrow Baseline Feature Tracking Techniques for Markerless AR Systems

Area of Science:

• Computer Science
• Computer Vision
• Human-Computer Interaction

Background:

• Augmented reality (AR) systems integrate real-world and computer-generated data.
• Accurate registration, aligning virtual objects with the real world, is a critical challenge limiting AR usability.
• Existing methods often rely on man-made markers or struggle with complex environments.

Purpose of the Study:

• To propose a novel, marker-less natural feature tracking based registration method for augmented reality applications.
• To enhance the usability and robustness of AR systems in diverse conditions.
• To enable accurate online pose tracking and augmentation.

Main Methods:

• A natural feature tracking approach using a reduced Scale-Invariant Feature Transform (SIFT) based augmented optical flow tracker.
• An adaptive classification tree based matching strategy for fast and accurate initialization.
• Utilizing arbitrary geometric shapes (planar, near planar, non-planar) for registration.

Main Results:

• The proposed method demonstrates simplicity and efficiency, eliminating the need for man-made markers.
• Robust performance is achieved even with occlusions and significant viewpoint changes.
• Fast and accurate initialization is possible even with large differences between initial and reference camera poses.

Conclusions:

• The novel registration method significantly enhances AR system usability and robustness.
• The marker-less approach is suitable for both indoor and outdoor AR applications.
• Experimental evaluations confirm the method's effectiveness for online pose tracking and augmentation.