Question about the accuracy of tag's pose

Question

Question about the accuracy of tag's pose

SonOfCoding opened this issue 5 years ago · comments

Hi, I am running the aruco_tracker.py. After getting my own kinect v1 's Intrinsic matrix which is
[[1.04847513e+03 0.00000000e+00 6.66118289e+02]
[0.00000000e+00 1.04687412e+03 5.19355741e+02]
[0.00000000e+00 0.00000000e+00 1.00000000e+00]]

I try to get tag's pose (RT, 33 matrix and 31 matrix), I put my camera right above the tag. I think the hight (z direction) is correct. but the x,y direction has some error,

I think the distance between the center of the tag and the center of my camera is nearly equal to zero. However, the result (tvec) shows -5cm and -4cm. is there a certain amount of error during detection? does the tvec need to do extra calculation? or is there something I can change in the code to improve the accuracy? Please give me a hint, thank you

Anirudh NJ · Answer 1 · Sun Mar 10 2019 01:04:31 GMT+0800 (China Standard Time)

Did you calibrate for your camera?

SonOfCoding · Answer 2 · Sun Mar 10 2019 08:11:29 GMT+0800 (China Standard Time)

yes, i did use my kinect v1 to collect chessboard photo and use your calibrate_camera.py to do the calibration. and I have changed w,h(7,6) in your code, because mine is 11,8. did I do something wrong?

Matteo Pantano · Answer 3 · Tue Mar 19 2019 20:18:07 GMT+0800 (China Standard Time)

I am having the same issue. I did calibrate the camera but still the pose estimation is wrong

Anirudh NJ · Answer 4 · Wed Mar 20 2019 03:41:14 GMT+0800 (China Standard Time)

@matteopantano @SonOfCoding unfortunately i dont have much time to debug and research about that now..please if you find the result submit a pull request

Matteo Pantano · Answer 5 · Wed Mar 20 2019 15:45:10 GMT+0800 (China Standard Time)

Issue found, the problem was on the camera resolution. OpenCV by default opens 4:3. However, calibration was done on 16:9. Calibration resolution and opencv resolution must match.