Lidar360 & Cloud Compare - Remove Noise

if we talk about noise real fair two points in a point cloud that stay that are in a three-dimensional position that it is not real compared to the original object that has been somehow scanned or reconstructed through structure from motion processing noise is it's a factor it's something that we should deal when we edit a point cloud because it adds points to a point cloud so it raised up the Gigabyte so the amount of data of that point cloud but if we need to process a point cloud in order to build a mesh if a point cloud is extremely noisy we we can expect to have a noisy mesh tool in this video I'd like to show you how lighter 360 and cloud compare deals with noise inside a point cloud basically they used the same algorithm that fits a plane moving around the point cloud it measures the distances between points inside or in near a reference point from that plane and then decides if a point is considered as noisy as noise or can be left in the point cloud so this is the algorithm that basically this is how that algorithm works and it works like that number one step which is not number one that I've written here the number one step it actually is defining how many points are gonna taken in account in order to generate a plane the algorithm works for each point of the point cloud so considering this light blue points this is the reference point for this calculation as you can see there are points here that if you if you consider this line as a side view plane there are the screen points that are placed well on this plane but there are these points these red points that are in a wrong position considered that this could be the surface of a road or the surface of a wall this is the surface and these are points that shouldn't be there number one step is defining how many points and you can do by choosing a sphere that moves around the point clouds and this sphere should have a radius that should be should be big enough in order to allows the sphere to contain a good numbers of point in order to fit a plane that for that particularly specifical part of that point cloud and for that points so you have to define the radius of the sphere or otherwise but that works just for cloud compare you can define how many numbers that are how many points sorry that are around that are near to the reference point should be taken an account for fitting a plane it's quite similar to the SOR filter and you're gonna use the neighbor nearest neighbor algorithm once you define how many points that should be taken in account the algorithm first of all fit a plane then it calculates the distance the the average distance for each of this point to that plane hopefully if everything is good this average distance should be extremely low the best scenario is would be if this average distance we're zero but of course will not be because there are noisy points that will make the this this average distance differs from be different from from zero the next step is calculating the standard deviation of these distances then the software will take a standard deviation multiplier and will calculate a threshold distance then eventually with this threshold distance calculated it will check each of the this point distances with the threshold distance if the distance for of each of this point is bigger than the threshold distance that point will be considered as noise and will be removed otherwise it will be keep it will be kept inside the point cloud so that's the algorithm but let's go inside lighter 360 and inside cloud compare to see how it works keep in mind that is not easy to remove noise with just a couple of clicks and we're just working on that this algorithm that I'm gonna show you noise it's kind of a tricky thing inside a point cloud and sometimes you will need to manually remove it with some other techniques that maybe I'll show you in next videos and you you cannot 100% rely on an automatic algorithm that's the point cloud that I'd like to do to show you that I'd like to work on for dealing with noise it's a little section of a much bigger point cloud and it came from structure from motion processing in ladder 360 the noise removal algorithm you can find it in data management that's the the the bathe the the basic heart of the latter 360 it's not an external module as static sees as the tip sticks and classified data management point cloud tools noise filter once you launch that process a new window pops up and you have to choose which of the point cloud you'd like to process I've got some other point clouds here because I already did some processing because I'd like to avoid my CPU to raise up in terms of load and so that it could you could ruin my recording so I'm gonna work on this point cloud here which is the original one and here you can choose between you've got two options here number one option is radius you can choose between the recommended radius sphere or you can enter a radius value here if you choose recommended radius fear the software is not gonna ask for a measure but it will calculate the the minimum radius sphere for that sphere to move around the point cloud in order to be able to fit almost everywhere inside a point cloud a plane and so applying this kind of algorithm if you define a radius you can end manually enter in numbers here and you you have the chance to choose how big you would like to have this fear for this processing of course the bigger the sphere the much the more the numbers inside the sphere but bear in mind that if you if you apply this algorithm to surfaces that are for example corners maybe it will not work so well the other options that you have to choose here is the multiplier of standard deviation and it's basically this this number so you have to choose how much aggressive you'd like to have this filter you'd like to this we'd like to miss fifty to be active if you raise up this number you will define a much higher threshold and so last few point last point will gonna be removed if you lower this number you're gonna define a lower threshold and so you can expect you have much more numbers much more points removed from this processing here you have to define what will be the name of the new light data failed that file that the software will save inside your hard drive and choose the folder and once you've done just hit OK and the process will start I'm not gonna hit okay because I'm not gonna I'd like to avoid my CPU to overload so it canceled I'm gonna move to cloud compare now and movie to cloud compare the algorithm is basically the same so you have to select the point cloud here that you'd like to treat that you like to deal with this algorithm and then you had to go here tools clean and the noise filter this window has got basically the same information plus one more that the lider tool has the light of 360 tool has got so here you have the chance to define a radius fear once you launch this algorithm the software automatically calculate the minimum radius fear that it will he had judged to be the best one for your sphere to move around the point cloud and this is basically the option that you have inside lighter 362 that was just saying recommended regice fear here you can define a new radius fear so you can change this radius in order to have a bigger or a smaller radius fear compares got the options to fit a plane not using a sphere for point amounts calculate the amount of points to take in account but it can use the nearest neighbor tool so you can define how many points you'd like to take an account that are close to the reference point in order to fit the best plane that will fit that points in my opinion the radius option works better and that's the the defined options the default options when you launch this tool but of course you can definitely you have definitely to change to try different settings in order to find the combination that works best for you then in this in this window you have the chance to define the maximum error basically that's the definition of the standard deviation multiplier you have the chance to choose a relative standard multiplier well that's not a relative standard multiplier yes that's a relative standard multiplier because for each of this fear the software will take an account a different standard multiplier of different standard deviation of course that will be different for from a surface white surface rather than a much more crispy one so in this case it will adjust the the threshold and will adjust the filter in order to give you the best result compared to the fact that the point cloud that you're working on could be could have differences inside it if you choose absolute this kind of calculation will gonna be fixed for every area every part of the point cloud that you are dealing with both liner 360 and cloud compare has got the options to check or uncheck the remove isolated points if a point has got around it four or less points it's it's consider considered by this algorithm as an isolated point that because yes it's true that to fit a plane you need at least three point but for this algorithm the minimum number of points is six and if it doesn't find at least six point around one specific point of the point club it considered that point as an isolated point checking this box will I will yes will allow the software to remove it even if it's not considered as a noisy point but that's not considered as a noisy point just because the software wasn't able to fit a plane in the specific part of the of the point cloud so once you've choose once you've chosen all this stuff just hit OK again I'm not clicking okay just clicking cancel because I've already did some reading do some some calculation and some processing here so let's go and check what happened with the default settings inside cloud compare which were basically using the recommended wedges fear which was the base which has got the same radius as lighter 360 could could calculate and using one as a threshold standard deviation multiplier so I'm gonna switch it on and I'm gonna switch on and off the original cloud as you can see noise has been removed of course something happened this this new point cloud is less noisy but I think there's a little more there's there's some noise some some noise some rumors some dirt left here especially in this part in this area but you forgot if I move here there's some noise left here then I moved to another settings where I used I basically I lower down the stand room the the multiplier of the standard deviation so I applied a much more aggressive filter and I left no I'm sorry I didn't do that in this case I I increase the size of the sphere so I used 50 centimeters sphere moving around this point cloud leaving the multiplier as it was and was one so look at what happens so I'm gonna switch off the this first process point cloud I'm gonna switch on and off the original one look at this part this angle alright as you can see I used a bigger sphere and when the sphere was applied in this area of course some points of the sphere inside the volume of the sphere where maybe points of this vertical world some other points were points of this flat surface some other point was point of the corner so the software had you to decide something and had to fit a plane because this algorithm works with playing fitting so maybe he had fit a plane in this direction and of course some points were under or above the plane and has been removed so the result in my opinion is absolutely not so good and I have a lot of point that has been removed here so the wall is it's not so good-looking so I think that increasing the size of the sphere is not a good idea unless you have a really flat surface and you can move around it without any problems if you got even little areas with curvature so end of course corners you should avoid using a big sphere let's go back to the let's go back to the the recommended Reggie radius sphere and let's reduce the multiplier of the standard deviation so I reduce it so I applied a much more aggressive filter see what happens okay some noise has been removed of course but as you can see there are really a few points points left in the in the point cloud compared to the original one so this is the point cloud the original one I'm gonna switch off and you can see that it's a little bit more scattered here there really a few points left so applying an aggressive filter could work fine if you have a lot of points if you have a super dense point cloud if your point cloud is not so dance maybe it could not be a good idea because the result could not be so so nice here I try to apply the here I try to applied instead of using a sphere I try to applied the nearest neighbor approach for fitting a plane but the results was not so good so this is using six nearest neighbor there are there is some noise here left here with 50 no is 50 nearest neighbor points and the still noise here and 50 nearest neighbor points reducing again the trash or multiply the standard deviation multiplied again the the point cloud is much more scattered and than the original one I'd like to say I'd like you to approach also this thing bear in mind that as you can see removing noise is not so easy but you need to understand if what you are gonna - what you're trying to remove it's noise or its outlier it's not so easy understanding when when a point that shouldn't be there it's a noisy point or it's an outlier and in my opinion these two tools works really well combined together look at this look at the look at this in cloud compare but I'm gonna switch to lighter 362 if you go here in the tool clean the number one tool that is listed here is as for our filter then there's the noise filter if we go here data management point cut tools remove outliers noise filter I think that this the fact that they are written in this way it's kind of a signal that you should apply first to remove outlier as our statistical outlier removal filter and then on the remaining point cloud the noise filter and I think that that's the best combination for for removing noise so if I apply first the SOR filter I'm not gonna I'm not gonna go deep in the settings because I did it in my previous video I apply the SOR filter and let's zoom I'm inside lighter 360 now let's do a little bit better here and focus your attention on this area of course I'm not gonna I'm not gonna move all the the noise but I'm gonna switch off the original data okay and moving this way I think that it's much more much more clear what happened okay as you can see here there are a lot of points there I was considering them as noise but actually they were much more as outliers and let's go let's move here these points if I switch off the original data these points are much more like outlier outliers rather than noisy points once I applied the statistical outlier removal filter I can then applied the noisy noise removal filters on the noise filter and hopefully I can have a much cleaner dense cloud so if I apply the noisy filter I have a much more scattered dense cloud and of course it depends on the settings that you choose but I think that the remaining points are much more solid in terms of definition in terms of of noise here I have a better definition of this kind of path if I switch on the original data and I switch off you can see that I'm sorry about that I move the view okay let's go back here switch on and off the original data this is the original data on and off as you can see there's much less noise after the processing but that noise the noise that I removed was a combination between outliers and noise so of course and just to to finish this this this video of course there are parts of this dense cloud and it could be on the dense cloud that you're working on working YouTube there are still parts that should need some actual actions but basically I think that in this case you need to know when an automatic filter will reach we will have the job done job done and when you should start to work manually on the point cloud maybe you should consider the fact that the particular area of that point cloud is not so good and you can remove it maybe because a structure from motion processing did not perform well because it was and was the word the word grass the word plants and we're scattered things on the ground otherwise you can you can admit that you need to manually work on the point cloud in order to remove each point that you consider a noise point or something that is called a dirt points something that is dirt that somehow ruins the the point cloud in the particular area dealing with noise dealing with liars it really it's really tricky I think that you shouldn't be stuck on the default settings you should try different settings maybe on a little part of the point cloud in order to have the processing have a quicker processing and you should you should always check the whole point cloud when you you define the settings that works fine for a particular area and you apply to the old point cloud in order to see if something if that if that settings work well for the old point cloud and just checking if if everything was good I hope that this video was useful you can check you can you can reach me out at www.mathantics.com thank you for watching gonna see you in the next one ciao

Channel: Metrical Talks

Views: 2,270

Keywords: noise, lidar360, noise filter, point cloud, 3d, greenvalley international, software, tutorial, how to, metrical talks

Id: VtA_bd51y-k

Length: 25min 54sec (1554 seconds)

Published: Fri May 15 2020