I. After checking the results by using 2 different "control points", the authors proposed value shows more stable results. The results generated by using 2 metric (...i.e., without/with gaussian opimization process) are same. Here are the value used as the "control points":
- The original one is [1 0 0; 0 1 0; 0 0 1; 0 0 0]
- The second one is creating by me which are the centers of each red markers in the Kinect frame
II. The transformation is currently not correct. I try to check the problem by starting from the most simple case - "two Kinect frames and mapping the same 4 points which are the center of 4 corresponding red markers"
- Here are the preliminary results:
 |
| Transformation result (marks in magenta & green ) by using my defined control points |
 |
| Transformation result (marks in magenta & green ) by using authors defined control points |
- Here are the possible issues that need to be checked:
1. Normalization of the coordinates of input points?
2. Camera intrinsic value is precise?
3. Image coordinate vs camera reference coordinate?
III. Search for some existing values of doing calibration on a Kinect, make sure mine is not so precise:
- Mine:
fx = 528.24, fy = 529.69, xc = 418.79, yc = 279.31
No comments:
Post a Comment