Agisoft Metashape - Some tie point tools

99% of the times that we use a structure from motion software and we upload sub images inside it for processing and reconstruct a 3d model we are using pictures that are not calibrated there are have been taken from normal standard cameras such as drones camera point-and-shoot or DSLR or mirrorless even if we have a big sensor a full-frame sensor these cameras are not calibrated these cameras are really different from cameras that were used to that are used for classic photogrammetric and that are that have a detailed calibration sheet that comes with them structure promotion software during the alignment processing do a really interesting thing that is the auto calibration or the images that you have processed that the software has processed auto calibration means that the software is able to reconstruct the physical parameters physical settings all the cameras and these are the focal and the principal distance the principal point coefficient of distortion radial and tangential distortion and this information are used by the software to define the position of the cameras in the 3d space and that position is determined by the alignment process and the definition of the tie points of the sparse point cloud in this video I'd like to show you this information well where you can find this information inside edges of meta shape and then I'd like to show you a couple more tools that are related to the type of the sparse point cloud of the spot or the sparse cloud so here I have left a couple of chunks here these chunks these two chunks basically refer to the same data set but in the in this first case I processed it I process the images with the default parameters in terms default settings in terms of features point for each images 40000 and maximum type point matches type point limits for each pair selection that is 4,000 in these other chunks in this other chunk and it that's the difference in terms of point cloud in these chunks I I reset kind of reset all the filters so I ask the software not to filter anything in terms of features point detection and in terms of point matching between a couple of images when the alignment is done you can go here in tools camera calibration and this panel gives you a really interesting information about your camera and in terms of speaking about cameras I refer to images in this video I'd like to focus your attention to the very first two tabs initial and adjust it maybe I think that I will go through this panel much more detail in another video because there are a lot of things to say about the camera calibration but here let me be extremely quick and just showing you how to read this panel in this Venus in this area these are the kind of type camera type that is related to sensors sensor type and focal length if you should if you try - men - if you try to work with picture pictures there have been shot by different cameras or by the same cameras but with a zoom lens and you changed the the focal length so you move the zoom during viewer during your acquisition here you will have a number of types that are related to all the combination that you you had during you are during your shooting in the field in this case and I have to say for the majority of my work I try to stay fixed with just one type so that's just one combination of focal sense of focal length and sensor type and this is this type is telling me that I have 23 images this is the sensor sites and this is the focal length here you gather for each of this type you add the camera type in this case it's a frame camera but you can also have fisheye vertical or cylindrical cameras this is the pixel size this is the real focal length and here you got the information about the camera calibration initial is what the software had in terms of information at the very beginning of the alignment process so here you go out the focal length the the principal distance it must be said principal distance here we got k1 k2 k3 k4 these are the settings the parameter the coefficients for the radial distortion that CX and CY this is the position of the principal point this is p1 p2 p1 between these are the coefficients for the tangential distortion which is far less it is less important than the radial distortion all these settings all this information extract except for the focal length for this principal distance r0 the software know anything about this camera but at the end of the alignment all this information has been filled with a number which is different from zero the principal distance I think that it's the same stayed the same yes yes it's the same but all these other features have changed so the software was able to calculate all these information all the settings during the alignment process and now he has it has some information about what are the real well not the real because you do not have a calibration but it's much more detailed now the information about the camera are much more detailed because the software was able to calculate them and now it use it uses them for the alignment process if you somehow have a camera with a calibration sheet you know for sure all this information you can upload them so you can go here load and you can upload a txt file or and another file that is able to tell the software listen this camera as this precise information in terms of principle point radial coefficient coefficient of distortion and principle distance in this case I do not have them and for the vast majority of my work I do my works I do not have a calibration sheet but it may happen that you have really precise information about that so you can upload them so this is the camera calibration tool and just to show you what happens when the alignment has finished you can read all this information there if you'd like to access to have access to the tie points tools you can go here tools and tie points you have three options you add three tools well actually the tools are just you this is a tools that is this is a tool that allows you to check what happened between couple of emails between pair of the images so let's start with view matches view matches when you when you click on view matches this new window will pop up and this are here listed these are all the pictures of the data set and here in this menu you can access to all the pictures of the data set so you need to check by pairs so you need to check a picture here which is maybe we can start with the second one 5 1 3 and then by clicking on one of these pictures that are listed here you can check what are the matches between these two pictures as you can see this is a pair selection check so you just have to select a pair of images and you can check what happened between them in terms of matching points so let me check this maybe this one ok here here there are listed here are listed some information and here's the name of the pictures that you are checking with this one here these number is the total matching point beat matching points between these two pictures and among these matching points 2132 these are the numbers of matching points that have been considered valid 1663 and are colored in blue in this in this area where you have the two pictures and then there are some more pitch with some more points which are 469 that have been considered invalid so they are matching points but the software but the software decided that they didn't have a solid information they they weren't so solid in order to be part of the sparse point cloud so they have been discarded and that's it you can check every pair of image that you'd like by selecting the fixed one here and check-in clicking by the the others here in them in this list let me let me see let me let me check one thing this is the chunk where I am working in the chunk that I processed using zero and zero in terms of features point detection and matching point between Percel action so here as you can see I have met a number of matching points between a couple of images that are far more bigger than the four thousand then I have by default when I launch for a very first time the alignment settings the alignment images settings if I move let me close this one if I move to the high one to this one where I used four thousand as limit in terms of matching points between couple of images and I go back to here view matches as I can see there are no pair of images that have more than four thousand total matching points what actually the maximum number is this one and it's a little bit less than three thousand so that's that have been limited by the settings that I used for the alignment process that's very important this our last point but I think there are much more stable and much more reliable than the other the in this case we have 18,000 points for the sparse point cloud here we got 123 thousand points for this passed point cloud let's go back to tools tie points build point cloud basically that's the same tools in terms of settings and as you can see for the the menu that you have accessible here than the alignment tool but this is available only if you have information of the camera calibration so in this case we have camera calibration information because the software has finished an alignment process and have calibrated as calibrated all the cameras but if you have external calibration external information in terms of calibration sheet very precise you can upload them and you can build a point cloud with that information in order to reprocess them we're much more accurate information in terms of coefficient of radial distortion focal length principal distance and principal point basically that's the same thing if I run the build point cloud now I guess I will add this more or less the same result as I'm as I can as you can see here with the very first step of alignment so let's cancel this one and let's go to let's go through the the last tool here in time points which is thin point cloud in thin point cloud you can somehow go back to a sparse point cloud which has much less point because you will work the software will work in the pair selection mode and will recalculate the point cloud in order to have for each pair selection for each couple of images that I've been there will be check together a maximum number of matching points which is 4000 in this case if I click OK I will expect to have in output a sparse point cloud with a number of points that more or less will be really close to this one let me do that I think it will take really short time let me check click ok ok it was really really fast as you can see the number of points for the sparse point cloud have changed has changed now I have 17,000 points for this part cloud here and this number is I think that is quite close to these 18,000 points here from that that is being calculating from this from these settings inside this chunk if I switch between 1 and 2 as you can see the points are not the same because of course they cannot be the same I think that they are much they are a little bit better here in this in this chunk when I you where I use this default settings but more or less these are the same point and I can see less clearly here the subject the object of this data acquisition which is this roundabout but I can I can see that even if I process that the process the sparse point cloud and thin it with these tools so that's it for this video I think that knowing what happens inside the alignment process it's really important because the alignment process of the images of the data set it's the very first step of destructive promotion process and if the first step if the first results are not so solid every other step of the process will be not so solid - so I hope that you enjoyed this video I think that the view matches tool is really interesting for checking out what happens in terms of pair selection and definition the matching points of the one cloud between two images that have been checked together and I hope you enjoyed it if you if you'd like to share it I would really appreciate you can check me out at www.att.com/biz ow

Channel: Metrical Talks

Views: 3,128

Rating: 4.818182 out of 5

Keywords: agisoft, agisoft metashape, metashape, agisoft metashape pro, metashape pro, tie point, tie points, tie points tool, tool, tools, settings, software, tutorial, structure from motion, sfm, photogrammetry, how to, metrical talks

Id: kQGBVe646ck

Length: 18min 5sec (1085 seconds)

Published: Tue Jun 02 2020