Arbitrary viewpoint video synthesis from multiple uncalibrated cameras
read more
Citations
Systems and methods for the autonomous production of videos from multi-sensored data
Identification of 3d objects from multiple silhouettes using quadtrees/octrees.
Virtual Viewpoint Replay for a Soccer Match by View Interpolation From Multiple Cameras
BISi: a blended interaction space
Personalized production of basketball videos from multi-sensored data under limited display resolution
References
A versatile camera calibration technique for high-accuracy 3D machine vision metrology using off-the-shelf TV cameras and lenses
A versatile camera calibration technique for high-accuracy 3D machine vision metrology using off-the-shelf TV cameras and lenses
Light field rendering
A volumetric method for building complex models from range images
The lumigraph
Related Papers (5)
Appearance-based virtual view generation from multicamera videos captured in the 3-D room
Frequently Asked Questions (21)
Q2. What is the SS method for generating a floor plane?
Since the floor plane is removed at the step where the silhouette image is made, the SS method only provides the 3-D shape of an object without its background.
Q3. What is the advantage of image-based virtual view synthesis?
The precise and dense correspondences make it possible to generate virtual views at arbitrary viewpoints without losing pixels even in partially occluded regions.
Q4. What is the effect of the geometrical settings of the base cameras?
Since the PGS is defined by the basis cameras, the geometrical settings of the base cameras affects the results obtained by the proposed method.
Q5. What is the way to reconstruct a 3-D shape model from multiple-view images?
Reconstructing a 3-D shape model from multiple-view images requires a relationship between the 3-D coordinate of the object scene and the 2-D coordinate of the camera-image plane.
Q6. What is the way to represent a 3-D shape?
Although the visual hull reconstructed from silhouette images cannot represent an actual 3-D shape, the visual hull can be used as an approximation of the actual 3-D shape in some cases, such as IBVH [12] and the method presented in this paper.
Q7. Why does the error in 3-D shape reconstruction affect the quality of the generated images?
Because IBR is essentially based on 2-D image processing (cut, warp, paste, etc.), the errors in 3-D shape reconstruction do not affect the quality of the generated images as much as they do for the model-based rendering method.
Q8. How do the authors select the basis cameras?
The authors select two basis cameras so that the angle of the viewing direction between the basis cameras is close to 90 to make the axes , and almost perpendicular to each other.
Q9. How can the PGS be used to reconstructed objects?
By applying the PGS to the similar virtual view synthesis technique, the strong camera calibration required in conventional work can be avoided.
Q10. What is the voxel density in the PGS of the objective area?
Since the field of view of the cameras used in this experiment is less than 10 , the voxel density in the PGS of the objective area is roughly homogeneous.
Q11. How can the point be projected onto every input view image?
Since the coordinate of a point in the PGS is fixed, the point can be projected onto every input view image with the fundamental matrices in the same way as stated before.
Q12. Why is the 3-D coordinate in a PGS dependent on the camera coordinates?
Because the 3-D coordinate in a PGS is dependently defined from the camera-image coordinates, the 3-D position of the sample points does not have to be measured.
Q13. What is the method for reconstructing a 3-D shape model in a large target space?
In this section, the authors proposea method for reconstructing precise 3-D shape models in a large target space, which involves dividing the target space intoseveral small subcells and reconstructing a 3-D shape model for every cell.
Q14. What is the SS method used to determine the shape of the object?
In the conventional SS method, each voxel in a certain Euclidean space is projected onto every silhouette image with projection matrices (which are calculated by accurately calibrating every camera [2], [14]) to check whether it is included in the object region.
Q15. How do the authors obtain the fundamental matrices between the cameras?
The fundamental matrices between the cameras are obtained by putting a checkerboard pattern at various heights, as depicted in Fig. 10, so that the image feature points can be distributed in the objective space.
Q16. How many cameras can be mounted on the surface of a hemisphere?
if a number of cameras are mounted on the surface of a hemisphere enclosing the target space and any three of them form a triangle effectively, the virtual viewpoint can be moved freely all around the half sphere.
Q17. How many image feature points are extracted from the two basis cameras?
From such images, about 50 image feature points are extracted, and then the same feature points extracted in the other cameras are manually corresponded.
Q18. Why is the distance between the object and the camera relatively large?
Although the use of affine transform is not perspectively correct, the authors ignore such perspective errors because the distance between the object and the camera is relatively large in the present experiment.
Q19. How is the 3-D position of the viewpoint of the first basis camera determined?
4. Then, the pixel position on the interpolated view image for the vertex is calculated by the following equation:(5)Next, visibility of the vertex from each reference viewpoints is checked by comparing the depth from the reference viewpointto the vertex with the depth value in the depth image at the reference viewpoint.
Q20. What is the order in which the existence of a voxel is checked?
In this process, the order in which the existence of a voxel is checked is important for reducing the computational complexity, because the cost of computing the projection of a voxel onto an image is not the same for all the images in the proposed scheme.
Q21. How many cells are used to synthesize a 3-D shape model?
even when the target space is very large, e.g., a soccer field or an American football field, the authors can synthesize arbitrary view images by dividing the whole target space into several cells and reconstructing a 3-D shape model in each cell separately.