[vtkusers] 3d reconstruction

[celeste gonzalez]

Hi Wes,

Thanks for your help. I am using the transform determined at the previous
stage as an initial transform on the currente stage. For example, I am
registering the data set in 0 grades and 10 grades, then apply
the transformation to the images at 10 grades and with a new point cloud (20
grades) I use the ICP algorithm again, that is, my sequence of registrations
is 2 to 1, 3 to 2, 4 to 3, etc..

I got a perfect alignment until 90 grades. I've tried with many objects
but I always have the same result, the alignment at 100 grades is
erroneous...


On Fri, Jul 23, 2010 at 7:43 PM, Wes Turner <wes.turner at kitware.com> wrote:

> I'm not sure I understand exactly the sequence of registrations, but I can
> make a quick guess.  Most likely you are sequentially registering each data
> set to the same reference i.e. 2 to 1, 3 to 1, 4 to 1, etc. instead of 2 to
> 1, 3 to 2, 4 to 3, etc.  Unfortunately, ICP can easily fall into a local
> minimum and fail to converge if the initial positions of the point clouds
> are too dissimilar. An easy fix would be to use the transform determined at
> the previous stage as an initial transform of the current stage.  That would
> keep the registration errors within about 10 degrees and should allow
> everything to works as expected.
>
> - Wes
>
>   On Fri, Jul 23, 2010 at 7:21 PM, celeste gonzalez <celgon23 at gmail.com>wrote:
>
>>   Hi everybody,
>>
>> I want to reconstruct a 3D object via registration of stereo range data. I
>> am using vtkIterativeClosestPointTransform to align individual point sets
>> with respect to each other.
>>
>>
>> The sequence consists of 36 images pairs constituting a full rotation of
>> the object. I am using two consecutive points cloud with this code and then
>> I transform the source points by the matrix solution.
>>
>> Initially, I took the first pair of point cloud  then I use the point set
>> that was aligned and a new point cloud that I wish to align
>>
>> The problem is when the rotation angle it’s over 90 grades, that is, from
>> the tenth picture the alignment fails.
>>
>>
>>
>> This is part of the code I am using now.
>>
>>
>>
>>       //setup ICP transform
>>
>>       icp = vtkSmartPointer<vtkIterativeClosestPointTransform>::New();
>>
>>       icp->SetSource(newPolydata);   // newPolydata
>>
>>       icp->SetTarget(referencePolydata);   //  referencePolydata
>>
>>       icp->GetLandmarkTransform()->SetModeToRigidBody();
>>
>>       icp->SetMaximumNumberOfLandmarks((int)ceil(newPolydata->
>> GetNumberOfPoints()*0.9));
>>
>>
>>
>>       icp->SetMaximumNumberOfIterations(100);
>>
>>       icp->SetCheckMeanDistance(1);
>>
>>       icp->SetMaximumMeanDistance(0.001);
>>
>>       icp->StartByMatchingCentroidsOn();
>>
>>
>>
>>       icp->Modified();
>>
>>       icp->Update();
>>
>>
>>
>>       vtkSmartPointer<vtkMatrix4x4> M = icp->GetMatrix();
>>
>>       cout << "The resulting matrix is: " << *M << cout;
>>
>>
>>
>>       vtkSmartPointer<vtkTransformPolyDataFilter> Transf =
>>
>>       vtkSmartPointer<vtkTransformPolyDataFilter>::New();
>>
>>       Transf->SetInput(newPolydata);
>>
>>       Transf->SetTransform(icp);
>>
>>       Transf->Update();
>>
>>
>>
>> Thanks in advance
>>
>> _______________________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at
>> http://www.kitware.com/opensource/opensource.html
>>
>> Please keep messages on-topic and check the VTK FAQ at:
>> http://www.vtk.org/Wiki/VTK_FAQ
>>
>> Follow this link to subscribe/unsubscribe:
>> http://www.vtk.org/mailman/listinfo/vtkusers
>>
>>
>
>
> --
> Wesley D. Turner, Ph.D.
> Kitware, Inc.
> Technical Leader
> 28 Corporate Drive
> Clifton Park, NY 12065-8662
> Phone: 518-881-4920
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.vtk.org/pipermail/vtkusers/attachments/20100723/13069e6d/attachment.htm>