Eye vs Camera: Why it's nearly impossible to film an eclipse

in 2019 i tried to film a lunar eclipse with an interesting hdr time-lapse technique but the hdr wasn't quite deep enough to actually see the eclipsed red moon and the mount was shaking and the clouds rolled in and then i ran out of space on the sd card so it went really well so i waited for three years i bought some new equipment i made a plan i spent a week obsessively watching cloud maps across the state and then when the time came drove out somewhere that i hoped it would be dark and cloud free set up all my equipment and started rolling unfortunately it wasn't actually cloud free but i think that the footage that i got is still pretty cool this is a time lapse of the spring 2022 lunar eclipse and although it has a few problems notably some sketchy stabilization and an abundance of clouds it's a lot better than my previous attempt let me explain the motion of the moon the earth shadow and the background field of stars here because it's more counter-intuitive than you might expect and i want to talk about what makes this two-camera hdr process different from the vast majority of eclipse time lapses that you'll probably see [Music] during a total lunar eclipse the moon passes behind earth's shadow relative to the sun so the moon is no longer in direct sunlight i don't think that's going to surprise any viewers of the channel but this means that the moon is then illuminated only by light that scatters through earth's atmosphere so basically if you were on the moon looking at earth during a lunar eclipse you would see like a ring of sunset all the way around earth that's really the only light that reaches the moon which is why the moon looks red if you look at this as an outside observer you can imagine that as earth moves around the sun its shadow is going to point in different directions all the time and then the moon orbiting earth faster could pass through this shadow which would give us a picture like this where relative to faraway stars earth's total shadow or umbra would move from west to east at about one degree per day because it has to go 360 degrees about every 365 days and the moon would move west to east at about 12 degrees per day weirdly enough that's not actually what i filmed aside from tilts because orbits are all a little bit crooked i actually saw earth's shadow move overall from east to west relative to the star field i was pretty confused by this when i started analyzing the footage but then i realized i was filming from a moving vantage point i was not the outside observer looking down on the solar system from above if you are that outside observer then we expect a simple model where the umbra is going to move from right to left and the moon is going to move from right to left much faster and the moon's going to pass behind the umbra but if you're on earth's surface then you also are moving and over the course of one night you actually shift and you're looking at the moon from a different direction this time lapse took like five hours to film so in the early evening i was basically looking at the moon from this angle and then after a few hours you know past midnight i was looking at the moon from a completely different direction so relative to the background stars the moon actually appears to move specifically it looks like it's moving from east to west it looks like it's moving backwards turns out that this effect is actually much faster than earth's movement around the sun but angularly speaking much slower than the moon's orbit around earth which means that the moon still looks like it's going in the correct direction across the frame but it looks like it's going too slow and earth's shadow resolved at the distance of the moon not at a semi-infinite distance like the rest of the stars actually appears to reverse direction so with a whole bunch of canceling effects here earth's shadow actually appears to move east to west relative to the stars while the moon itself seems to move west to east it's a real mess but it results in a pretty cool view but there's more to this time lapse the thing that i was really trying to accomplish was not looking at the relative motion of earth's umbra and the star field i just thought that was cool and i wanted to open with it the goal of this time lapse was actually to create a more realistic view of an eclipse and it required a surprising amount of camera foolery so i have long been frustrated by seeing time lapses of eclipses solar and lunar where the camera adjusts halfway through so it looks like once the moon is fully behind the earth's shadow and it turns red it actually gets brighter and that's not what happens it might look that way to a camera or to your eye because your brain sort of auto corrects for how bright the environment is but in reality that is the darkest part of the eclipse because the moon's not being lit by sunlight it's being lit by sort of the aggregated halo of sunset that is all the way around earth and uh that's not very much light that is scattered through earth's atmosphere that said tonight i am planning on fixing that so i have here two different cameras that are both pointed at the moon and both set up to run for about five hours so we should get the penumbral eclipse and then the actual eclipse and then the outgoing penumbral eclipse and this camera right here is set up to take a bracket of three pictures and properly expose sort of the regular moon like the normal bright moon that you're used to this camera is set up to properly record the eclipsed moon the red moon and none of these settings are going to change so i'm going to combine these i'm going to have four images for every frame of this time lapse and i'm going to combine these in such a way that the moon actually gets dimmer the whole time it's becoming eclipsed the first time i tried this a few years ago i had a single camera on a homemade barn door tracker and i was taking the widest bracket that the camera would allow which basically means that every time that the shutter was pulled by the intervalometer the camera took three pictures at three different exposures one the dimmest exposure was at one thousandth of a second and this was able to properly capture the sunlit moon this is what you'd use to take a picture of like the full moon and then there was an intermediate frame and then the brightest frame was actually holding the shutter open for a full 15th of a second not a thousandth of a second and this is dramatically more light and you can see that when you're looking at one of the images of the full moon it's just totally bleached out but when you look at one of the images of the eclipsed moon the eclipsed moon is actually too dark so we need something even more separated and for that i needed a second camera fast forward a few years and i now have a second camera and i bought a second second well a second lens which was also a second hand lens but i got another lens that was exactly the same so that i could have two cameras with although they're not the same sensor they have the same size sensor so that i could mount them together pointed at the moon and have them spin on a mount that was better than my barn door tracker and try to get two frames that had completely different exposures a minute ago i said that i was actually taking four frames a three frame bracket on one camera and a single frame on the second camera but for the final composition i actually only ended up using two of those frames the 1000th of a second same as the first time but for the bright image i'm using a 2 second exposure so this is a 2 000 times difference in light between these two pictures so they're going to be really dramatically different if i represented this completely linearly then i mean you you wouldn't see the eclipsed moon like the brightest pixel on the eclipsed moon is going to be something like a thousand times dimmer than the brightest pixel on the regular moon so if you're looking at the regular moon on your phone screen or your computer screen that resolves colors between 0 and 255 then literally every pixel of the eclipsed moon is going to be below one and you're just gonna get a field of black on that screen so that's not very interesting it's also not what your eye perceives when you look at the night sky your eye does not see things the same way that a camera does your eye and brain together do a fantastic job of selectively making parts of your vision brighter and darker to maximize contrast across the field now it's very possible to see stars right next to a really bright moon with your eye but it's not possible by taking a single exposure in a camera likewise halfway into a lunar eclipse your eye can totally see that half of the moon is starting to look red even when it hasn't been fully covered up yet and that's because your eye is perceiving the faintest of contrast that is actually below the dynamic range threshold of the camera in this image that i've spliced together you can see that half of the moon is white half the moon is red and then there's this sort of muted gray region in the middle that your eye just sort of glosses over and it ends up showing you the two high contrast regions so when i wanted to shoot a time lapse where you could see both halves of the moon the whole time and the moon got progressively darker as it got more eclipsed i artificially amplified the dark footage the red moon footage just by taking longer exposure and then in processing was very careful that i never allowed that footage to get brighter than the light moon footage with the fully sunlit moon the key is that both cameras were not automatically adjusting their exposures both cameras were fixed the entire time they were processed individually with processing that stays the same for every single frame of the time lapse and then in the most literal sense of the word they were added together if you're familiar with the concept of transfer functions in image processing i think that the human eye perceives a basically logarithmic transfer function of the incident light coming in but here i'm cheating and i'm basically just connecting two straight lines this means that all the colors of gray that the camera was able to perceive between zero and one thousandth of a second of brightness are represented by the values zero to 127 on your screen and then all of the colors between a thousandth of a second and two seconds are represented by the values 127 to 256 on your screen of course there's a bit more complexity to the whole image processing and i do apply some local contrast and some sharpening that fudges these numbers a little bit it's still very close and as the moon gets eclipsed it gets progressively darker and then as the moon is uneclipsed it gets progressively brighter which is what i really wanted to see unfortunately there is still a problem if you look at any one of these frames there's this sort of muted gray region in between the light area and the dark area and this region basically doesn't have any contrast it's just like a 127 gray now earlier i said that this is the bit that your eye kind of glosses over but that's sort of a simplification this only exists in the image because i cheated with the transfer function with enough technology we really should be able to do better but i would have needed a lot more pictures granted i did actually record four pictures in between thousands of a second and two seconds but i only ended up using the two extremes which actually makes this band a little bit thicker because i had an issue with stabilization i was expecting that if i had two cameras mounted to a solid stick that as that stick moved and tracked the sky that i would have no problems i'd be able to take the time lapse from one and the time lapse and the other and stack them straight on top each other unfortunately it didn't happen both cameras were sort of jittering back and forth relative to each other i think that some of this might have come from the lenses not actually motion of the camera bodies but regardless it made for really jittery footage turns out that when you're trying to composite images on top of each other even extremely slight deviations are super noticeable i mean if the whole moon moved up and down by two pixels for a frame it really wouldn't matter but if half of the moon moves up and down for a frame and the other half doesn't then your brain gets really confused because the moon is like no longer shaped like the moon you got both halves doing different things that said there are a lot of great automatic stabilization tools nowadays but for images that are mostly black with just sensor noise for the algorithm to grab onto maybe a few stars that the algorithm isn't optimized to see and otherwise just a smooth disk davinci did not handle it very well i even tried the cloud tracker which ended up tracking a cloud instead of the moon i don't know what i was expecting so instead i did it by hand filming the time lapse took 20 seconds per frame and i was basically just sitting there watching the camera reading a paper and eating pretzels stabilizing the entire thing frame by frame and starting over a few times took significantly longer i cannot understate the tedium here i went through multiple iterations of getting the rotational offset between the cameras correct trying to lock a star in the background on the over exposed footage by literally moving forward one frame at a time and moving the star back into the same position and then tracking the moon's edge when the stars weren't visible assuming that the moon was moving in a straight line at a constant speed which it actually doesn't because of all the parallax stuff i talked about earlier and after all that it's still very far from perfect the correct way to do this would be to write a piece of code that could actually register features on the moon's surface or detect the edge of the moon or even latch onto individual stars but with the clouds moving in and out of frame and stuff like that it was gonna end up being really difficult to automate that so i didn't and i did it by hand and it's not as good as i would like it to be but it's of course i wrote the code i've been watching this time lapse i don't know i probably watched it 100 times while editing the rest of this video and every single time i looked at it i'm like it's not good enough this is the real problem with the channel having like hundreds of thousands of subscribers now that my standards go up and that i don't actually end up publishing anything in a reasonable time frame to describe this as succinctly as possible over the last week i combined a matlab script that i wrote years ago that d flickers night time time lapses and a matlab script that i wrote years ago that stacks atomic resolution microscope images into this horrible inefficient monstrosity that de-jitters this one particular time lapse and it only works on this time lapse because so many things are hard coded it reads in 21 images at a time filters them with a really aggressive unsharp mask to find the edges and mostly ignore the clouds it stacks the adjacent frames on top of each other and uses a ridiculously inefficient sort of convolution method to find out how much each adjacent image needs to be shifted in order to line up with the others after it decides which frames actually have pictures of the moon in them and not just the clouds it uses all of these adjacent frame offsets to run a fit for x and y and then it moves the frame saves it stuffing my hard drive with hundreds of unnecessarily high bit depth lossless tips and then it works i really hope that you like it it's also why i'm probably the last one publishing my eclipse footage like weeks after the fact because i thought i was going to be able to process it a lot faster i guess the moral of the story here is that sometimes it is really hard to take an idea and even with a lot of planning and you know all the equipment that you think that you need it can be really difficult to make that idea into a thing i really try to show the failures on this channel whenever i can i mean it's in the title card plan a always goes up in flames and i think that in general especially online there can be a perception that everything always works exactly the way it's presented in a video the first time when that is literally never the case i think the channel is finally large enough that we should be able to start something that is self-sustaining and for a while i have wanted to start a subreddit specifically dedicated to learning from failures i might post there but i don't want to make it about the channel i want to learn from you i want to see the projects that you're working on i want to see your iterations and the stuff that you've learned by going through those iterations by making things that didn't work and trying it again and making something that was better the format that i'm shooting for is at least two pictures where you show something that didn't work and then show something that does work and in the comments say what the difference was so not just saying i made it better but see how you made it better so that everybody can learn from each other's mistakes this is my example post i guess when i was trying to make optical mirror coatings on printed parts i coded a whole bunch of benchies some of them worked and some of them didn't i talked about this iteration quite a bit in a couple videos ago i had a couple questions about what types of projects i was looking for what things could be submitted and i think i mean the answer is basically anything someone specifically brought up uh baking bread as an example i think that would be perfect i can show you downstairs i have pages of notes from like five or six years ago when i was trying to come up with a pumpernickel recipe that didn't have the consistency of a brick so i mean whether you want to call it the scientific method or the artistic process or iterative improvement as a most vague term it's all the same thing it's how people work on stuff and i think it's really cool so thanks for watching i hope that you enjoyed the time lapse and i really cannot wait to learn from you guys and see all the projects that you're working on so when the title card is done playing in like 25 seconds go check out the new subreddit i'll see you there [Music] you

Channel: AlphaPhoenix

Views: 84,714

Keywords:

Id: 6R-iMmcQ5aI

Length: 19min 50sec (1190 seconds)

Published: Sun Jun 12 2022