Abstract Augmented reality technology is a new technology developed on the basis of virtual reality technology, which is a new development of virtual reality technology. The key of augmented reality technology is to integrate the 3D virtual objects generated by computer with the objects in the real scene, so as to supplement and enhance the real world information. Augmented reality technology is involved in many fields of computer science. Most of the early researches focus on the problem of 3D registration. In recent years, with the deepening of the research, how to improve the visual consistency of the augmented reality system, and provide a more realistic and soft interface of virtual reality, has become a new research hotspot. In order to improve the sense of reality in augmented reality system for the purpose of synthetic images, proposes a new occlusion algorithm, this algorithm avoids the usual way in the past, but the operation mechanism of Open based on Inventor, the design of dynamic EMO node, and the virtual scene is divided into hierarchical modeling, while the use of moving target under complex background the extraction technology and Alpha channel technology, to achieve multi-level virtual real scene mutual occlusion. Since the algorithm does not need to model the real environment in advance, so a lot of preliminary work is reduced, and the versatility and portability of the system are improved. More importantly, the algorithm for non rigid objects, can also be a good completion of the virtual and real mutual occlusion.
Introduction
In Augmented Reality (AR), achieving "visual consistency" is the ultimate goal for immersion. While early research focused heavily on registration—ensuring virtual objects remain anchored to the physical world—a more complex challenge is mutual occlusion.
If a real person walks between the camera and a virtual object, the person should naturally block the object. Traditionally, solving this requires "prior knowledge," such as a detailed 3D scan or model of the real environment. This requirement is computationally expensive and significantly reduces the portability of AR systems. This paper proposes a novel occlusion algorithm that functions dynamically in real-time without the need for pre-modeling the environment.
Methodology and System Architecture
The authors developed their solution on the Illusion platform, utilizing DirectShow for video capture and Open Inventor for 3D scene management.
The proposed approach consists of four key components:
- Hierarchical Virtual Modeling: The virtual scene is divided into depth-based units (VP1, VP2, VP3). This allows the system to "sandwich" real-world objects between different layers of the virtual world.
- Moving Object Extraction: The system identifies moving targets in the real-world video feed (the "Moving Object Area" or MOA). This allows the system to identify potential occluders without a pre-built map.
- Alpha Channel Technology: Once a moving object is extracted, the system generates an Alpha channel (a transparency mask) based on the object's contour.
- Moving Obstruct Nodes (MON): These custom nodes are placed within the virtual hierarchy of the Open Inventor scene graph. When a real object moves into a position where it should block a virtual one, the corresponding MON is activated, using the Alpha mask to render real-world pixels in front of the virtual ones.
Key Innovations: The Scene Graph "Sandwich"
A significant insight of this research is treating the real-world video feed as a standard node within the 3D scene graph. By placing the BackgroundVideo node at the root of the graph, the system can manipulate the rendering order effectively.
By treating the scene graph as a layered structure, the authors can handle non-rigid objects like human limbs or flowing clothing. Because the system relies on real-time contour extraction rather than rigid 3D models, it can successfully handle occlusion for complex shapes that are notoriously difficult to model accurately in real-time.
Results and Discussion
The proposed algorithm offers several advantages over traditional modeling-based approaches:
- Improved Portability: Because the system doesn't require a preliminary modeling phase of the environment, it can be deployed in unknown rooms immediately.
- Visual Consistency: The algorithm achieves a "soft interface" and high sense of reality, successfully managing multi-layer occlusion where virtual objects can be both in front of and behind different real-world elements simultaneously.
- Efficiency: By using "dormant" MON nodes that only activate when occlusion is detected, the system manages computational load effectively.
Limitations and Future Work
Despite its successes, the system has limitations. It focuses primarily on moving object extraction; it is unclear how it handles static real-world occluders (like a stationary pillar) if they were not moving when the system initialized. Additionally, the visual quality depends entirely on the accuracy of the moving target extraction under complex backgrounds. Future research should include quantitative performance metrics, such as frames per second and latency, to further validate the system's utility in high-speed AR applications.
Summary updated