REAL-TIME OCCLUSION BETWEEN REAL AND DIGITAL OBJECTS IN AUGMENTED REALITY

Open Access
- Author:
- Jain, Pranav
- Area of Honors:
- Computer Science
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Conrad S Tucker, Thesis Supervisor
Rebecca Jane Passonneau, Thesis Honors Advisor - Keywords:
- Augmented Reality
Computer Vision
Machine Learning
Model Fitting
Occlusion - Abstract:
- In this thesis, a mobile-based, augmented reality (AR) method is presented that is capable of accurate occlusion between digital and real-world objects in real-time. Occlusion is the process of hiding or showing virtual objects behind physical ones. Existing approaches for addressing occlusion in AR applications, typically require the use of markers, depth sensors, and power- ful computers like laptops and desktops. Furthermore, real-world environments are cluttered and contain motion artifacts which result in occlusion errors and improperly rendered virtual objects, relative to the real world environment. These occlusion errors can lead users to have an incorrect perception of the environment around them while using an AR application, namely not knowing a real-world object is present. Effective methods to solve these occlusion errors are usually computa- tionally complex, requiring the use of high-end laptops or desktops to run. Moving the technology to mobile-based AR environments is necessary to reduce the cost and complexity of these tech- nologies. This thesis presents a method that brings real virtual objects into a similar coordinate system so that virtual objects do not obscure nearby real-world objects in an AR environment. This method captures and processes visual data in real-time, allowing the method to be used in a variety of non-static environments and scenarios. The results of the case study showed that this new method could reduce compute complexity, maintain high frame rates to run real-time, and maintain occlusion efficacy. Keywords: Augmented Reality, Computer Vision, Machine Learning, Model Fitting, Occlusion