Met with Mikki at 10 AM this morning to capture checkerboard patterns for the various GoPro settings. She has a GoPro 4, 5, and 7 to test with and each have different setting options.
We jumped in the SCUBA pool and took video for each setting. After drying off, we took a look at what we captured. Somewhat surprisingly, the "linear" mode was the worst and "wide" was the best. Linear mode automatically rectifies images in air and (I think) wide mode does nothing or very little. But underwater, the index of refraction change warps linear mode dramatically. So badly that we thought that we had mislabeled our test sets. So we waded into the pool and took video while we transitioned from air to water to confirm the effects. Indeed, linear mode would be a terrible option underwater.
We found that the wide mode was wrong by about 10% at the corners and that the error isn't across the width. (So at 1/4 the frame width the error is less than 5%). Practically speaking this means that the integrated error across the frame is very likely less than 5%. There are other noise sources that can be larger including single-camera parallax, the shark/fish/ray not being perfectly straight, relative pointing angles, and so on. In the long run, I still want to be able to correct that error, but it's not the tall pole to hammer on. So we'll advise the users to use wide mode in the mean time and think about how to improve that as part of our 6-month goal.
Speaking of which, that will be the next thing we do when we meet. We need a road map that we can work towards. However, I'll be out for the next two weeks on vacation, so there's going to be a hiatus.
We jumped in the SCUBA pool and took video for each setting. After drying off, we took a look at what we captured. Somewhat surprisingly, the "linear" mode was the worst and "wide" was the best. Linear mode automatically rectifies images in air and (I think) wide mode does nothing or very little. But underwater, the index of refraction change warps linear mode dramatically. So badly that we thought that we had mislabeled our test sets. So we waded into the pool and took video while we transitioned from air to water to confirm the effects. Indeed, linear mode would be a terrible option underwater.
We found that the wide mode was wrong by about 10% at the corners and that the error isn't across the width. (So at 1/4 the frame width the error is less than 5%). Practically speaking this means that the integrated error across the frame is very likely less than 5%. There are other noise sources that can be larger including single-camera parallax, the shark/fish/ray not being perfectly straight, relative pointing angles, and so on. In the long run, I still want to be able to correct that error, but it's not the tall pole to hammer on. So we'll advise the users to use wide mode in the mean time and think about how to improve that as part of our 6-month goal.
Speaking of which, that will be the next thing we do when we meet. We need a road map that we can work towards. However, I'll be out for the next two weeks on vacation, so there's going to be a hiatus.
Comments