3D Reconstruction from Depth and Stereo Images for Augmented Reality Application - Antoine Mischler
- © Copyright??
In this thesis, we introduce a framework for
augmented reality applications that require depth information in
real-time. We combine time-of-flight (TOF) range imaging based on the
photonic mixing device principle with stereo vision. The data is
merged at interactive frame rates, providing depth maps that are more
reliable and accurate than the ones from either of the sensors. The
confidence of the TOF depth data is estimated. The resulting
confidence map is used in two ways. First, with the TOF depth data we
initialize and constrain the disparity range of the stereo algorithm.
Second, we segment the images according to the color information and
depth confidence information. This segmentation produces
depth-continuous data areas. It allows to adapt the correlation
window to the depth data. The first part helps avoiding unnecessary
computations and disambiguating disparities and the second part
increases the stability near depth discontinuities. This algorithm
produces accurate and reliable depth maps and we use them in augmented
reality setups. We show that this range data can be used to handle
accurate visual occlusions and physical interactions between virtual
objects and the real world.