Agisoft Metashape - Advanced settings for image alignment

when you ask a structure from motion software for image alignment you may check or map by saying you may decide some settings you may change some settings and these settings will somehow define the result of the alignment these settings are the number of key points or features points that the software is able to define and to take in account for each image of the data set and the other setting is the number the maximum number of tie point that the software is able to define from the pair selection mode in the in the second step of the alignment processing in this video i'd like to show you how to change the settings inside adisoft meta shape and what are the differences between a processing with a certain amount of points key points and features point and compare with another processing with a much higher number of points in terms of features point and tie points so let's go to workflow align photos and we need to go to the advanced settings so let's close the tab that is related to general settings and go to advanced settings we have five options here but we will see much details these two the very first two key points limit key point limit and tie point limit so what is key point limit key point limit is the maximum number of key points or feature point that the software software will consider for each image of the data set step number one of the alignment images on the align images processing is detecting features point so the software will check every image of the data set and for each of this image will will define the features point this number is the maximum number of features point that the software will take for each image and will bring to the next stage next stage is the tie point is the the the pair selection mode in order to transform as many key points to tie point checking images in a pair selection mode this number um doesn't mean that the software will start from zero and then if when he it reaches 40 000 it will stop it will go through all the key points all the features point that it is able to define for each image and then it will keep just the 40 000 best points for each of these key points the software um assigned to each of them kind of a square and it is able with this square it is able to to know which are the best key points among all the key points for each images so it will go through all of them and i will keep just the 40 000 best tie point limit applies this kind of a filtering in the pair selection mode so for each of the pair that are checked by the software in the second stage of the alignment process the software will take in account just the the first 4 000 points 4 000 tie points maybe the software is able to define a higher number of key points of us i'm sorry of tie point or matching point starting from the key points but it will consider just the best ones so the best the four thousand best ones that doesn't mean this doesn't mean that the sparse point cloud will have 4 000 points of course this is the maximum number for a pair selection mode you have we you will definitely have a lot of pair to be selected and to be checked one with each other and the number of the points of the sparse point cloud will definitely be higher but this is the limit the maximum limit for the pair selection mode very very quickly you can check or uncheck the guided image matching if you check it the software the process will uh would benefit from this check or unchecked boxes if you have high resolution images and i'd like to add also that the images should have been should have been shot with the with with a big sensor sites so in this case i have pictures that have been taken from a drone dji phantom 4 pro it's a quite uh it's uh it's not a big sensor and there are not so high resolution images so i do not have basically i will not have any benefit from this guided image matching box checked or not adaptive camera model fitting is checked by default if you uncheck it the software will calculate just some of the calibration parameters that i will go through in next videos uh that is able to calculate in particular if you uncheck the adaptive camera model fitting the software will will process will calculate the focal length the principal point position and three coefficient for the radial distortion and two coefficient for the tangential distortion if you check this the software will will be able to calculate all of the factors that it can take in account for the camera calibration mode this will this will not uh transform a poor data set uh result in a high standard quality result of course not but it will help the process for a better camera calibration processing if you if you put zero here and if you put zero of course you can put any numbers if you want and there's no there's no point in uh in selecting a higher number of tie points than a key pointer because of course tie points are related to key points um if you if you insert zero here and here you're not gonna limit the software but you're gonna tell the software do whatever you want and do not apply any filter so keep any key point limit taking account any key point limit for each images do not filter them and keep any tie point link any tie point sorry keep any tie points that you will be able to find in the pair selection mode so there is no limit inserting 0 here writing typing 0 here is like saying there's no limit that's not always a good idea but i'll go through a couple of processing that i did in order to show you the results so i'm hit cancel and i'll go through these four chunks where i uh where i processed the software processed four different alignment with four different settings in terms of key point limit and tie point limit let me go back here for each of these four set i keep all this box checked of course not with that current alignment i choose the high accuracy high accuracy and i keep adaptive camera model fitting checked so number one for processing number one i used a key point limit of 10 000 and a tie point limit of 1 000 for each of these four chunks i maintain a scale factor of of 10 between key points and tie points so this is the sparse point cloud of this processing and as you can see this punch point cloud is made of 4 and 400 points in terms of processing time this parts point cloud took let me see 28 seconds 28 seconds to to be processed plus 6 seconds 30 seconds to be processed so let's move to the default settings which are 40 000 for the key point limit and 4 000 for the tie point limits the sparse point cloud is made by 18 and 600 points there are a lot more than the previous one and you can better see what's the the subject of this uh this processing destructor from motion processing this is one of the pictures this is a roundabout and you can better see it here while here it's not so clear well i know that because i i did the i took the pictures on the field uh but it's not in the field but it's not so so easy to understand here it's much more defined how much did it take one minute and 25 seconds one minute more than the previous one then here i processed um i choose 100 000 points for the key points limit and 10 000. for the tie point limit so now my thai points limit is the key point limits of the very first processing that i show you you can see here that the sparse point cloud is much more defined you can better see what's on the ground number of points 20 51 and 500 and how much did it take here five minutes and 10 seconds four minutes more or less four minutes more than the default settings processing and then in the end i'd like to show you what happened if i type 0 for the keypoint limits and 0 for the tie point limits as you can see the sparse point cloud here is it's much more dense it's not dense it's it's much there are much more points here one one two three two zero six and you can better see there's a lot of points here you can barely see the sign here on the uh this is the the sign on the ground on the and you can you can barely see heat here them here the this point is there are much more points here of course how much did it take five minutes and 40 seconds as you can see it took just four seconds more than the ha this this uh with then the processing with the settings basically because in this case the software didn't apply any filter applying a filter will takes time so the software will take time for applying filter in order to keep just the best point in this case he processed all the images it processed all the pair selections and it just she just showed me everything that he had calculated without any filtering if i zoom in if i move around this point cloud what i can see is um is that this point cloud yes it's nice if i see from far from from quite from above and not if i do not zoom in sorry about that but if i move it around i can see that this point cloud is a little bit noisy there's noise here i'm looking at this point cloud at kind of road level there's no it sees of course i'm i'm i have i've inverted um i was in the wrong in the wrong view uh there's noise and you can see this noise there's points that instead of being on the on the surface this is the the road pavement there are some points that are above and some points that are below and that's because the software didn't apply it didn't apply any filtering filtering which means uh limiting the number of key points and number of features point helps the software for the best results in terms of uh sparse point cloud and of reliable reliability of this point so my suggestion is try to work with the default parameters default settings if you look through the agisoft user manual you can see that edgesoft itself says that the best settings are 40 000 for the key points and 4 000 for the tie points of course these settings this is not low you can change uh you can change whatever you want and you can try different settings uh i suggest you to try different settings to see what happens in the software but don't be tempted in pushing the limit too far and type in zero in both of that boxes because what you may have in output is yes a sparse point cloud with a lot of point but maybe that point are not so stable are not so good as the points that you may have in output choosing and limiting the number of key points and the number of tie points consider the fact that the sparse point cloud the tie points will be the basis for next processing especially for the dense point cloud processing and if you have points that are not so good points of the sparse point cloud that are not so good you may have a not so good dance point cloud so consider uh keeping the number of points lower applying some filter limiting the number of key points and the number of tie points in order to have the best output for the sparse point cloud i hope that this video was useful feel free to share it feel free to connect with me www.metrical.com thank you so much for watching see you in the next one ciao

Channel: Metrical Talks

Views: 13,671

Rating: 4.9039998 out of 5

Keywords: agisoft, agisoft metashape, metashape, metashape pro, agisoft metashape pro, image alignment, features point, tie point, point limit, sparse cloud, matching points, metrical talks, tutorial, software, photogrammetric, photogrammetry, structure from motion, sfm, how to

Id: AWso2m2FPkA

Length: 16min 43sec (1003 seconds)

Published: Mon May 25 2020