1 | initial version |
The problem was with the calibration of the camera itself. I was assuming, in the OpenCV calibration, that the aspect ratio of the focal length was the same for x and y, and that the camera primary point was in the center. Checking the documentation of the RealSense I realized that neither os those assumptions was correct. I forgot about that, and extract the calibration intrinsic parameters directly from the camera API. With that fixed, and the chessboard improved, I got the results of the picture below.
As you can see the results now are much closer. Thank you so much for the help!
2 | No.2 Revision |
The problem was with the calibration of the camera itself. I was assuming, in the OpenCV calibration, that the aspect ratio of the focal length was the same for x and y, and that the camera primary point was in the center. Checking the documentation of the RealSense I realized that neither os those assumptions was correct. I forgot about that, and extract the calibration intrinsic parameters directly from the camera API. With that fixed, and the chessboard improved, I got the results of the picture below.below. All of them with 20 poses.
As you can see the results now are much closer. Thank you so much for the help!