SPARSE
3D RECONSTRUCTION OF A SCENE FROM VIDEO (M.S.
Thesis)
An Overview of Sparse 3D Reconstruction
The purpose of sparse 3D reconstruction
is, given an image sequence, to determine the 3D location of some sparsely
selected points in the scene. The reconstruction process has mainly three
parts, such as feature detection and matching, epipolar geometry estimation and
metric reconstruction of the points selected as features.
•
Input: Image sequence viewing a scene
•
Output: Point cloud (3D Location of selected points)
Simulations
·
Reconstruction
results for Krimm
sequence (Obtained by using 8 frames from the original video at equal intervals
of 10 frames). Click on the images to view the videos.
Original
video Reconstructed scene
·
Reconstruction
results for Cevo
sequence (Obtained by using 10 frames from the original video at equal
intervals of 50 frames). Click on the images to view the videos.
Original video Reconstructed
scene