ret
3.5233523642607523
camera_matrix
[[976.95273119   0.         660.46033674]
 [  0.         983.93214293 265.09898857]
 [  0.           0.           1.        ]]
dist
[[ 0.00786213 -1.08703858 -0.03323499 -0.00704253  3.32886628]]
rvecs
[[[-0.15258765]
  [ 0.05568271]
  [-0.00297466]]

 [[-0.1975074 ]
  [ 0.18257839]
  [ 0.01421636]]

 [[-0.63995505]
  [ 0.01228971]
  [ 0.01489291]]

 [[ 0.17243489]
  [ 0.10097482]
  [-0.02850808]]

 [[-0.16032189]
  [-0.4157807 ]
  [ 0.07556575]]

 [[-0.0815965 ]
  [ 0.57730837]
  [-0.01252756]]]
tvecs
[[[-0.1267968 ]
  [-0.04263508]
  [ 0.31507263]]

 [[-0.11701824]
  [ 0.00296928]
  [ 0.5963867 ]]

 [[-0.1100471 ]
  [-0.03971764]
  [ 0.3491404 ]]

 [[-0.12381456]
  [-0.02769804]
  [ 0.28588741]]

 [[-0.10678357]
  [-0.04163386]
  [ 0.30308073]]

 [[-0.09151737]
  [-0.03837504]
  [ 0.44047188]]]

I got 6 vectors in rvecs and tvecs as I used 6 images of checkerboard with different tilting angles.
I also understand I need to convert this 3×1 rvec to 3×3 rotation matrix by

# Convert rotation vector to rotation matrix
R, _ = cv2.Rodrigues(rvec)

My current project is detecting an object (using YOLO and single camera) and calculate & display the distance between camera and the object by comparing actual length of the object and length in pixel image.

Here are my questions:

1. As one point could be located further than the another point from the camera, and this causes the shorter distance in pixel. To estimate/ calculate more percise & accurate distance, do I need to convert two points in image coordinate (in pixel) into world coordinate/ camera coordinate (in m)?

2. If world coordinate is required, which ‘rotation vector’ and ‘translation vector’ from the camera calibration should I use (or should I use the average of them)?

3. When undistorting points, do I use same camera matrix for P=, or have to get new camera matrix by using “cv2.getOptimalNewCameraMatrix”?

# Undistort the pixel coordinates
undistorted_point = cv2.undistortPoints(np.expand_dims(pixel_coords, axis=0), K, dist, P=K)