Recognition with in noisy data

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Recognition with in noisy data

krips89


Hi,

I am trying to register point cloud from a mesh (eg. a ply/obj file) to a noisy set of point cloud (for example the data from a reconstruction of that mesh including the points in the background)

I am looking for some suggestions for choosing right set of Algorithms and parameters. I want to have the following properties for the registration/recognition: 
1. Scale invariant - the reconstructed noisy data can be of different scale from the input mesh.
2. Robust to noise - The reconstructed point cloud can be quite noisy.

I am uploading a sample set of data (Model - Final.obj, reconstructed scene - dense.cmvs.0.ply) in this email. 

Can you guys help me in focusing on some correct set of algorithms available in PCL? I am considering both Local and Global pipeline. Recognition in a structured point cloud from kinect was easy with the 3D recognition app (global pipeline), but with noisy scene, it failed.

Also, how do you determine the correct parameters for the method shown in the correspondence-grouping tutorial? Can  you tell me the specific parameters for my data? and the way you inferred them so that I can solve similar problem in the future myself? 

Thanks in advance!

--

Kripasindhu Sarkar




--

Kripasindhu Sarkar


_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users

dense.cmvs.0.ply (3M) Download Attachment
Final.obj (271K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Recognition with in noisy data

andersgb1
There's just one problem with your whole use case: none of the algorithms you are using are scale-invariant. I know there's a method for computing a relative transformation with scale in PCL, but the methods are not using this one by default...

On Mon, Aug 10, 2015 at 6:49 PM, Kripasindhu Sarkar <[hidden email]> wrote:


Hi,

I am trying to register point cloud from a mesh (eg. a ply/obj file) to a noisy set of point cloud (for example the data from a reconstruction of that mesh including the points in the background)

I am looking for some suggestions for choosing right set of Algorithms and parameters. I want to have the following properties for the registration/recognition: 
1. Scale invariant - the reconstructed noisy data can be of different scale from the input mesh.
2. Robust to noise - The reconstructed point cloud can be quite noisy.

I am uploading a sample set of data (Model - Final.obj, reconstructed scene - dense.cmvs.0.ply) in this email. 

Can you guys help me in focusing on some correct set of algorithms available in PCL? I am considering both Local and Global pipeline. Recognition in a structured point cloud from kinect was easy with the 3D recognition app (global pipeline), but with noisy scene, it failed.

Also, how do you determine the correct parameters for the method shown in the correspondence-grouping tutorial? Can  you tell me the specific parameters for my data? and the way you inferred them so that I can solve similar problem in the future myself? 

Thanks in advance!

--

Kripasindhu Sarkar




--

Kripasindhu Sarkar


_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: Recognition with in noisy data

krips89

Hi Anders,
Can you tell me the name of the method? I will take a look in its documentation.

On Aug 11, 2015 8:25 AM, "Anders Glent Buch" <[hidden email]> wrote:
There's just one problem with your whole use case: none of the algorithms you are using are scale-invariant. I know there's a method for computing a relative transformation with scale in PCL, but the methods are not using this one by default...

On Mon, Aug 10, 2015 at 6:49 PM, Kripasindhu Sarkar <[hidden email]> wrote:


Hi,

I am trying to register point cloud from a mesh (eg. a ply/obj file) to a noisy set of point cloud (for example the data from a reconstruction of that mesh including the points in the background)

I am looking for some suggestions for choosing right set of Algorithms and parameters. I want to have the following properties for the registration/recognition: 
1. Scale invariant - the reconstructed noisy data can be of different scale from the input mesh.
2. Robust to noise - The reconstructed point cloud can be quite noisy.

I am uploading a sample set of data (Model - Final.obj, reconstructed scene - dense.cmvs.0.ply) in this email. 

Can you guys help me in focusing on some correct set of algorithms available in PCL? I am considering both Local and Global pipeline. Recognition in a structured point cloud from kinect was easy with the 3D recognition app (global pipeline), but with noisy scene, it failed.

Also, how do you determine the correct parameters for the method shown in the correspondence-grouping tutorial? Can  you tell me the specific parameters for my data? and the way you inferred them so that I can solve similar problem in the future myself? 

Thanks in advance!

--

Kripasindhu Sarkar




--

Kripasindhu Sarkar


_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users


_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: Recognition with in noisy data

andersgb1

On Tue, Aug 11, 2015 at 8:30 AM, Kripasindhu Sarkar <[hidden email]> wrote:

Hi Anders,
Can you tell me the name of the method? I will take a look in its documentation.

On Aug 11, 2015 8:25 AM, "Anders Glent Buch" <[hidden email]> wrote:
There's just one problem with your whole use case: none of the algorithms you are using are scale-invariant. I know there's a method for computing a relative transformation with scale in PCL, but the methods are not using this one by default...

On Mon, Aug 10, 2015 at 6:49 PM, Kripasindhu Sarkar <[hidden email]> wrote:


Hi,

I am trying to register point cloud from a mesh (eg. a ply/obj file) to a noisy set of point cloud (for example the data from a reconstruction of that mesh including the points in the background)

I am looking for some suggestions for choosing right set of Algorithms and parameters. I want to have the following properties for the registration/recognition: 
1. Scale invariant - the reconstructed noisy data can be of different scale from the input mesh.
2. Robust to noise - The reconstructed point cloud can be quite noisy.

I am uploading a sample set of data (Model - Final.obj, reconstructed scene - dense.cmvs.0.ply) in this email. 

Can you guys help me in focusing on some correct set of algorithms available in PCL? I am considering both Local and Global pipeline. Recognition in a structured point cloud from kinect was easy with the 3D recognition app (global pipeline), but with noisy scene, it failed.

Also, how do you determine the correct parameters for the method shown in the correspondence-grouping tutorial? Can  you tell me the specific parameters for my data? and the way you inferred them so that I can solve similar problem in the future myself? 

Thanks in advance!

--

Kripasindhu Sarkar




--

Kripasindhu Sarkar


_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users


_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users