With the fast-paced innovation of technology nowadays, we can create things with the tips of our fingers. Through many applications like COLMAP, we can reconstruct objects by simpling inputting the images to their pipeline. For this project, we will dissect the methods. The main goal is to be able to combine all the knowledge we have learned from the course and put it into practice by reconstructing a simple scene.
The first part is to calibrate the camera which will capture the scene subject for reconstruction. Through a virtual checkerboard, we will obtain the intrinsic parameters which will then be used for the next tasks.
Size of the images:
- 1080 pixel image: 190mm x 190mm
- 720 pixel image: 130mm x 130mm Resolution of the image in pixels: 2048 x 1536 pixels
Internal parameters:
| Number of Images | 7 |
|---|---|
| Scale factor (f_x) | 1756.19431759707 |
| Scale factor (f_y) | 1733.56613931276 |
| Principal coordinates | 776.117376270950, 1091.94520035728 |
| s | -6.59288912302190 |
Are the pixels of your camera square?
Judging from the scale factor, we see a small deviation of approximately 1% difference. Because of this little difference, we can neglect this value and see that the pixels can be considered square.
Which is the degree of coincidence between the principal point and the center of the image plane?
The principal point coordinates are 776 and 1092 while the theoretical center of the image is 768 and 1024. Before concluding the that camera is properly calibrated, the theoretical and computed central point of the images should be closed. From the obtained values, we observe an approximately 3% deviation from the expected values for both the width and height.
Are the axes of the image plane orthogonal?
The orthogonality of the axes is obtained through the formula f_y = s * tan(x) where x is the angle from the respective axes that conform to the pixel. Through simple algebra of isolating x, we get a value of -0.34°. The value is so small that we can consider the values are almost perpendicular although noise is present.
Small picture:
| Number of Images | 7 |
|---|---|
| Scale factor (f_x) | 1767.05482387525 |
| Scale factor (f_y) | 1774.40669930832 |
| Principal coordinates | 686.384320925834,1060.17096039263 |
| S | 5.29999517265277 |
Radial distortion
In general, it is acceptable to neglect radial distortion as observed from the figure above.
Basing from the deviation, it is better to use the bigger checkerboard. Now, our camera is calibrated! Time to get the perfect combination of descriptor and detector for the best pair of images.