search

computer-vision / takahashi2012cvpr

An Implementation of Takahashi, Nobuhara and Matsuyama "A New Mirror-based Camera Pose Estimation Using an Orthogonality Constraint" presented at CVPR 2012
https://computer-vision.github.io/takahashi2012cvpr/
Other
26 stars 11 forks source link

Improve accuracy #4

Open TNA8 opened 2 years ago

TNA8 commented 2 years ago

Thanks for your great work.

I run your matlab version and got results. Before that I calculated ground truth matrix(R and T) for laptop built-in camera. Here is the result.

screenshot_5

Is there any way to improve the accuracy? If I use n-points version with 5 mirrors instead of 3-points version, could it be improved?

Thanks in advance.

nbhr commented 2 years ago

Hi, thank you for your interest in our work. We definitely recommend the n-points version. In general, use of as many as possible points results in a better (i.e., accurate and robust) pose estimatation.

haijing1995 commented 2 years ago

Thanks for your great work.

I run your matlab version and got results. Before that I calculated ground truth matrix(R and T) for laptop built-in camera. Here is the result.

screenshot_5

Is there any way to improve the accuracy? If I use n-points version with 5 mirrors instead of 3-points version, could it be improved?

Thanks in advance.

hello, Can you tell me how you calculated ground truth matrix(R and T) for laptop built-in camera?

TNA8 commented 2 years ago

Manually calculate the camera position relative to the monitor's left top corner. Normally the x-axis of camera coordinate system is opposite to the x-axis of monitor coordinate system, which results in opposite z-axis.

TNA8 commented 2 years ago

Hi, thank you for your interest in our work. We definitely recommend the n-points version. In general, use of as many as possible points results in a better (i.e., accurate and robust) pose estimation.

screenshot_25