I notice a lot of the complaints are about “sliding” in scans - either freehand or on turntable. I also notice the Mini at least seems to have accelerometers onboard (detects angle of dangle during calibration). In both freehand and turntable scenarios, it seems like the accels could be used to at least minimise these problems (if they’re not already). I’m just curious about the inner workings and how it’s currently done.
In turntable scan mode, it is clear that there should be only one axis moving at once, this should vastly reduce the tracking algorithm work, and errors, if it knew up front that this was a static-camera situation (different scan mode?). It seems that this isn’t done currently as we can easily get a multi-axis spiral when tracking features get reduced.
When scanning freehand or on turntable, it should be quite possible to use some estimations from the accels to try to corroborate the motion of the unit with the changing features in front of it. i.e. if I measure 2 seconds linear acceleration in the +X direction, and I see 300mm feature movement in the Y direction, something’s not right.
I’m curious how the system fundamentally works in the background.
The use of accelerometers in 3D scanners has been around for a while, but is apparently not as simple as one would hope. The technique is referred to as SLAM (Simultaneous Localization And Mapping) and it is not cheap. The least expensive 3D scanner I could find that incorporates this capability has a list price of $49,000.
Had the same idea when was first scanning with Mini. It seems it doesn’t incorporate this feature in any way since when I tried to scan a Hubsan Zino 2 quadrocopter arm with it, and the scanner got to the point, where that thin part of it was the only thing visible to the scanner, the scan began to quickly “flee” like that arm was about meter wide. It was obvious there is no inertia sensor involved.
Though unsure does it have those sensors at all even I too thought “hey, it have at least some sensors”, when was calibrating it, since it is not a smartphone that have a whole mess of those.
Considering Jeff said up there about “that would cost a lot” - can’t agree really, since the price is what you put on your product depending on what the competitors do, it doesn’t necessary have anything with actual production or even development cost. Like resin 3D printers were 5K USD minimum for about 8 years ago even the technology was already at the level, when some were building one themselves (even I made one), and now they are started at 150 USD, that is even more than 10 times difference. Yeah, those ones from the past had an expensive projector involved, but even that didn’t couldn’t explain B9Creator price except for “try to find cheaper”.
So I agree with you that Revopoint could try to get inertia tracking involved in at least to a some small degree. Maybe I am wrong, but I think it is possible.
@JeffLindstrom SLAM doesn’t require accelerometers, but aggregation of that data too can be part of the process - plenty of SLAM systems running from webcams in my old robotics lab. All SLAM requires is measurement of distances (to map) and the ability to track where you are in space, and record it…which is pretty much what these scanners do in reverse anyway. You can do it with a lidar line scanner in 2D. The accels just augment that info to give you greater confidence. I think people tend to call is SLAM when it’s used for large area mapping vs small 3D scanning. But the principles are the same. But I wouldn’t be surprised to hear that the Revopoint scanners use some degree of accel input, maybe just not fast or sensitive enough to rely on it.
@Harh agreed. I suspect some element of it is incorporated, but all the difficult tracking points (round sharp edges, areas with no features) would benefit massively from understanding the scanner’s attitude in space. Your mobile phone can do it with ease with cheap and cheerful onboard sensors.
Just to return and add a little more… I think the accuracy and drift requirements needed to use an IMU as the base of the measurements (i.e. to give some “true location” would be painfully expensive. But as an amalgamation - relying on the structured light as the truth, with IMU as a sanity check, I think that’s a lot less demanding. eg “I’m sweeping round here” - the data from the scan tells me I’m rotating X degrees per second in the Y axis - does the accel agree? Or "I’ve lost track for this nanosecond - instead of binning data, I’ll use the output from the IMU to assume it looks like this, hopefully I’ll pick up the features again in half a second.
It don’t detect anything while calibration , the small white point row indicate the horizontal line while calibration . It is optical based calibration only . It really don’t matter how you holding the scanner or what angle the calibration board is , all what matters are the big and small dots while the scanner taking single pictures of each position to calculate the difference.
Gyroscope is used to calculate the object in space in the first frame after you click the start button , it locking the object into that space so when you lose tracking it knows where is the starting point .
And that is …
The issue some people have are purely software based and the problem is addressed already by the development team, it is a bug but it don’t show up on all systems , at least not on my no matter what device I am using , but some experience it . It actually losing the starting point position of the object in space what affect the frame alignment.
But I know what you mean by , it would be of course game changer but for that Revo would need to build a new scanner hardware as current devices are visual based scanners , and the tracking is purely based on the object features and it’s point of origin only , no matter what feature markers or regular .
Interesting! It didn’t seem to move the line when I was moving the calibration board, but did when I moved the unit. Maybe I was going nuts, it was late, that’s why I assumed it was accelerometer/IMU based.
I’m not sure how the gyro being used to calculate the first point helps, because on turntable tracking it won’t know what happened to the rotating part. Certainly those are my primary annoyances that could be solved by some simple assumption/checks. Of course it would add problems in other areas too!
If they have even a basic gyro right now, and flexibility over when to use it, it could make a huge difference now, even just to capture those impossible misalignments.
It is actually illusion , I was thinking the same in early begining , you can try yourself , just put steady your MINI on the table and in front the board , move it right up left down and you will see the horizontal line will move based on the vertical position of the small dots that go through the middle of the big dots .
As that line of small dots together with the big dots are important to be aligned for the pictures that are taken while calibration .
It calculates the position in space of the very first frame only , all other frames after are based on the objects features only , that’s really nothing . Once it lose the object visual feature it lose the tracking exactly as it track the horizontal level while calibration , everything visually based calculations .
Same way as face tracker cameras works .
They would need to change the hardware and the technology they are using , Revopoint specializing in 3D tracking cameras in general for industrial purposes as well .
Their new scanners are purely based on visual tracking only .
I don’t have any issues with tracking or whatever is happening lately with users of MINI , I do have most all editions including the newest and everything is working just fine , as long you don’t try to scan something what you don’t suppose to scan with . Mini after all was designed to scan very small objects with detailed surfaces, the FOV is too small to track large areas .
You see it is very interesting , you can scan a leather sofa pillow that have very fine leather pattern on it , but you can’t scan a coffee mug without markers… and that is all about visual tracking …
Not to mention the software ,
so all ideas of yours and others are great , but it calling for complete remake starting with the hardware and after the software .
However I believe the hardware is great , what is lacking support here is the software .
Looking for example the new dual axis turntable, the software do not even offer 20% of what the hardware can really do , what in my opinion is worthless.
But thanks to our smart community members that pushed it beyond and above to something completely new and so much better with the 2 new apps for it .
Right now there are of course limitations in the Revo Scan software algorithms that can optimal handle the hardware.
I know because I was working on MINI improvement with the team months ago to improve the total volume accuracy and volume resolution in the software .
The Hardware is capable for even better results but pushing the software in that direction resulting in corrupt algorithms and failure , so the improving process getting very slow literally by trial and error.
That is where this technology hit the bottom line at this moment but with not so bad results after all so far considering the price of the unit , but it gonna be improved as the time goes, thankfully Revopoint is not a reseller of other companies technology and can move it to the next level with the time , meaning you going to have improved device and software with the time .