3D Reconstruction and camera pose from video sequence using multi-dimensional descent
This paper aims to propose a novel and simple method for estimating 3D-point reconstruction and camera motion. Given a video sequence of a target object with a few feature-points tracked, the points' 3D-coordinates can be reconstructed along with the estimation of the camera's position and...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Published: |
Springer Verlag
2015
|
| Subjects: | |
| Online Access: | http://www.scopus.com/inward/record.url?partnerID=HzOxMe3b&scp=77951128832&origin=inward http://cmuir.cmu.ac.th/handle/6653943832/38579 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Chiang Mai University |
| Summary: | This paper aims to propose a novel and simple method for estimating 3D-point reconstruction and camera motion. Given a video sequence of a target object with a few feature-points tracked, the points' 3D-coordinates can be reconstructed along with the estimation of the camera's position and orientation in each frame. The proposed method is based on combining Powell's method using parabolic graph with the well-known Gradient Descent to guess the direction to estimate the unknown variables. The unknowns include six components for camera pose in each frame, one focal length, and three values for each point. Using this proposed method, the problem of missing points due to selfocclusion can be eliminated without using any other special strategies. A synthetic experiment shows accuracy of computing 3D-points and the camera pose in each frame. A real-world experiment from only one off-the-shelf digital camera is also shown to demonstrate the robustness of our approach. © 2010 Springer-Verlag Berlin Heidelberg. |
|---|