Multi-view video camera setups record many images that capture nearly the same scene at nearly the same instant in time. Neighboring images in a multi-video setup restrict the solution space between two images: correspondences between one pair of images must be in accordance with the correspondences to the neighboring images.
The concept of accordance or consistency for correspondences between three neighboring images can be employed in the estimation of dense optical flow and in the matching of sparse features between three images.
This work has been funded in parts by the ERC Grant #256941 `Reality CG` and the German Science Foundation, DFG MA2555/4-2.
Code and ResourcesThe code and the test sequences are for research purposes only. No commercial usage is allowed in any form. If you use this code for your publications, make sure to cite the corresponding papers. Start downloading the test sequences by clicking on the images.
- MATLAB implementaion of the Three Image Optical Flow, including feature matches
- MATLAB implementaion of the Three Image Feature Matching
- Synthtic stereo camera sequences with ground truth correspondences:
The ground truth motion between neighboring images is given in an .exr file.
Waving sequence (2 viewpoints, 24 frames): Stonemill sequence (2 viewpoints, 3 frames):
Correspondence and Depth-Image Based Rendering: a Hybrid Approach for Free-Viewpoint Video
in IEEE Trans. Circuits and Systems for Video Technology (T-CSVT), vol. 24, no. 6, pp. 942-951, June 2014.
Capture and Statistical Modeling of Arm-Muscle Deformations
in Computer Graphics Forum (Proc. of Eurographics EG), vol. 32, no. 2, pp. 285-294, May 2013.
Dense Correspondence Field Estimation from Multiple Images
PhD thesis, TU Braunschweig, June 2011.
Monsenstein und Vannerdat, ISBN 978-3-86991-339-1
Correspondence Estimation and Image Interpolation for Photo-Realistic Rendering
PhD thesis, TU Braunschweig, April 2011.
Robust Feature Matching in General Multi-Image Setups
in Journal of WSCG, vol. 19, pp. 1-8, February 2011.
Traditional optic flow algorithms rely on consecutive short-exposure images. In contrast, long-exposed images contain integrated motion information directly in form of motion blur. In this project, we use the additional information provided by a long exposure image to improve robustness and accuracy of motion field estimation. Furthermore, the long exposure image can be used to determine the moment of occlusion for the pixels in any of the short exposure images that are occluded or disoccluded.
This work has been funded by the German Science Foundation, DFG MA2555/4-1
The goal of this project is to develop algorithms in image space that allow photo-realistic editing of dynamic 3D scenes. Traditional 2D editing tools cannot be applied to 3D video as in addition to correspondences in time spatial correspondences are needed for consistent editing. In this project we analyze how to make use of the redundancy in multi-stereoscopic videos to compute robust and dense correspondence fields. these space-time correspondences can then be used to propagate changes applied to one frame consistently to all other frames in the video. Beside the transition of classical video editing tools we want to develop new tools specifically for 3D video content.
This project has been funded by ERC Grant #256941 `Reality CG` and the German Science Foundation, DFG MA2555/4-2.
We present a method for image interpolation which is able to create high-quality, perceptually convincing transitions between recorded images. By implementing concepts derived from human vision, the problem of a physically correct image interpolation is relaxed to an image interpolation that is perceived as physically correct by human observers. We find that it suffices to focus on exact edge correspondences, homogeneous regions and coherent motion to compute such solutions. In our user study we confirm the visual quality of the proposed image interpolation approach. We show how each aspect of our approach increases the perceived quality of the interpolation results, compare the results obtained by other methods and investigate the achieved quality for different types of scenes.
The Virtual Video Camera research project is aimed to provide algorithms for rendering free-viewpoint video from asynchronous camcorder captures. We want to record our multi-video data without the need of specialized hardware or intrusive setup procedures (e.g., waving calibration patterns).