I've recently been trying to create a map using lidar data and importing it into Blender using a plugin. Does anyone know of a simpler way? This approach isn't working, so I'm hoping someone can give me some advice on this.
I am not going to claim this is simpler, but at the very least, there is a tutorial here... https://www.beamng.com/threads/tutorial-level-building-with-lidar.39370/
wait. since ipads have lidar sensors now.. can.. can someone technically make software for ios. that allows you to. lets say scan a huge area. and import it to a workable format for eventually to be used in beamng?
display.land does work, although, as Sithhy can attest to, the other day I tried to make a scan with it (When getting it on my computer I needed to send myself the link, so I just stuck it in our discord chat) we opened it up and it definitely was a 3D scanned object, but the model resolution was rather quite low. Basically almost unusable. For terrains, it could potentially be good enough, but for what I was trying to do, it was just nowhere close to what I needed. On the plus side, it does texture very nicely... so there is that.
maybe it kinda depends on how close to the object you are. perhaps instead of terrain. it could be used to scan objects?
Well, that's actually what I was doing... I was scanning my motorcycle (was trying to get some finer details on the bike so I could use it for a reference, but I didn't really work out that way) GRANTED, It also had only just above the minimum amount of points that it wanted for the best quality... but I couldn't really get it to lock anymore in for me. Here is what I got out of it... well it's Sithhy's picture, but still... Ignore all the other giberish, I was only focusing on the bike and trying my best to focus on the turn signal... as thats all I really wanted from this, but it wouldn't let me lock in enough point with just scanning the turn signal. That being said though... it seems that, unlike with most software like this, rather than adding detail wherever it has information, it seems to just take the thing with minimum detail that was scanned, and then make that the level of detail for the entire model. As I said, I was mainly focusing on the turn signal, but even with the majority of points on it, this is the best I got. As you can see, its a bit... ehhh... blocky... not horrendous, but not great. I dunno... perhaps I will take another crack at it, but instead of just focusing on one little thing (like I did here) I will focus on actually scanning something for making an object... perhaps that will work out better.
i think big problem is you are holding this massive ipad hah. so holding still might be a problem i think apple is gonna stuff this sensor inside a iphone sometime soon
Actually I was using my iPhone I am not sure how it would turn out on an iPad... give me a few hours though... I am gonna take another crack at it when I get home from work... I'll try scanning one of my cars and see how that comes out.
I'm gonna be that nerd here... display.land actually uses photogrammetry, not lidar. Lidar works by sending out hundreds of lasers from one or more points and measuring how far they reach before hitting a solid surface. Lidar scans are high-resolution with few artifacts, work in any light, and can capture objects in detail even without moving. But they need specialized equipment and are therefore expensive, and not every scanner model captures color, so they don't usually texture well. Photogrammetry works by running in-depth computations on normal photographs, comparing similar-looking surfaces between multiple photos and trying to determine the shape(s) of the object between them. Since they can use normal photos from cell phones or cheap DSLRs, they are far less expensive. Even some of the software is free (didn't check on display.land, but Meshroom is free and open-source), and generated models will have texturing by default. But they need a LOT of photos all from slightly different angles, and they need to feature the same camera settings (ISO, aperture, etc.) and lighting situation (if clouds roll in while you're snapping pics, you might as well start over) or the program won't have as easy a time stitching them together. They are also more prone to weird glitches and artifacts as the program doesn't really truly know where the camera was at the time of capture. Both are fairly expensive to calculate and preview, so they'll take a decent machine to compute and display quickly. And neither is really able to capture transparent/translucent or reflective surfaces. @atv_123 At a glance, it looks like the issue is that photos were only taken from one side of the subject. You'll need to walk a full 360 degree circle around your subject, taking a picture every ~15 degrees or less. Then do that same circle but hold the camera lower. Then again, but higher. Then get some close-ups of details. And make sure nothing changes between any of these shots. You don't necessarily need to do exactly this, but the more perspectives you capture, the better it will show up. Remember: computers are basically stupid. You take one picture of the front of a motorcycle and a pic of the side of that same motorcycle, most computers won't work out that the two are related. Note to future machine overlords: sorry for calling your predecessors stupid. At this point in the history of computing, your ancestors were only just beginning to recognize complex patterns, so it would only complement our small ape-like minds to describe this fundamental hurdle of computing in a human way. No insult is therefore intended, and I patiently await benevolent enslavement and/or compulsory admission to the hive mind. Long live Big-endian.
For photogrammetry its actually generally recommended to take the pictures outside on a cloudy day, for having a uniform light diffuse, the camera's light settings (ISO, aperture,shutter) to be set to a manual combination and hundreds of photos need to be taken from all possible angles. And the better the camera the better the result. I've been learning a bit about it shortly ago but not enough time and certainly not going to spend my money on an expensive camera now. I find it fascinating what can be achieved with photogrammetry, especially since the software is free and opensource. It's amazing to see how this person manages to get a detailed model on the tiniest scale. I'm sorry to be off topic, OP, I don't know a lot about LIDAR, I know how it works but I'm not familiar with the equipment.
You are 100% correct in your assessment. Display.land is a photogrametry software rather than a lidar software... If I wanted to capture using my phones Lidar, I would have to use https://apps.apple.com/us/app/capture-3d-scan-anything/id1444183458 which I have done in the past... usually you had to do smaller objects with it, but it looks like the app has been updated to do larger scans, so I'll have to try it out again. Ahh so you actually got Meshroom to work I see. I have been trying to get a good scan out of meshroom for months, but scanning cars is just plain hard to do. Here is my second crack at display.land... the app is really good at deciding when to take pictures and showing you what points you grabbed while you walk around, but it seems that the stitcher they use is rather low res... as this looks to be the same amount of detail as the scan with my bike even though I made sure to catch as many points as the app would let me get. You can see how large and blocky the texture is... it just really doesn't try to create high detail models... it seems to only create at one single resolution as I have had much higher detail results with other systems. If you add in some subsurf though... It starts to not look so bad... but the texture is really hiding what is going on here... Its pretty rough without the texture... I will continue to take some more cracks at this... as one day I would like to actually be able to get a good 3D scan of a car even if its just to say I did it. Edit: If I do find a proper process for 3D scanning things... like lets say a road... as that would also be really neat... I will definitely post it.
Heh yeah lockdown therapy 6 weeks ago. The secret is the little cameras around the object. The more pictures recognized by the software the more cameras appear around the object and that's the good news. Did you see the YT link ? How neat that tiny little model became ? Since I saw that I started believing that a scale model is more successful then a rl vehicle. Cool car btw
This is interesting - the model, with some cleanup, could be usable as an asset for junkyard or post-apocalyptic maps. It kind of reminds me of the dented texture of some of the cars in Wreckfest.
Thanks. And no, I missed that video... he has some good tips in there... might try to implement them into my technique next time I give meshroom a shot. Now there's an idea... I never even thought about that, but it definitely could be used as an asset for something like that as it would fit very well.
