Abstract
Progressing rapidly in recent years, augmented reality (AR) technologies have successfully found applications in engineering, marketing, entertainment, and medical industries. However, to render high-quality visualization contents in AR remains a challenging task with several technical barriers. Recent development of RGB-D cameras may provide an effective approach to solving those difficulties by capturing real-time depth data of a real scene. This research attempts to improve the effectiveness of product display in AR from multiple perspectives. First, a virtual model can be precisely positioned into a real scene according to the reference geometry selected by users from the scene, without use of markers. Mutual occlusions between real and virtual objects can be precisely estimated using the depth data. This information enables realistic interactions in an AR environment with intuitive rendering of both. In addition, a computational framework is developed to overcome data deterioration induced by random noises and missing pixels in the data captured by RGB-D cameras. The framework effectively reduces the jitter problem occurring along the occlusion boundaries. In addition, we implement a distributed AR system to realize the concept of mass virtual try-on. This system integrates cloud computing and parallel processing technologies to increase the computation efficiency involving in the try-on application. A use scenario of shoes try-on demonstrates the feasibility of the system. This work enhances the realistic extent of interactions between real and virtual contents by integrating RGB-D cameras into AR applications.