Wavy planes

Wavy planes

My first experiment with the Creative camera and Intel Perceptual Computing SDK was to put a giant planar surface about 12 inches from the camera and approximately fronto-parallel, and to dump out the 3D point cloud for the surface. I expected to see a noisy plane. But I get a surface that has a definite wave or bump in it. I checked with a level that my rigid board is really flat. I repeated the experiment with other flat boards, including one that is just entirely white, and I always get the same result: a wavy surface.

I've attached a couple of snapshots of what I am getting. I color the (X,Y,Z) points using the RGB camera image, and I use black if the location in the RGB image plane is out of bounds. That's why the points around the periphery are black.

Has anyone else observed this behavior?

Intel: Why does this happen and how can I remove this bias?

I am NOT computing the point cloud myself. I am obtaining the 3D point cloud directly from the SDK using PXCImage::COLOR_FORMAT_VERTICES.

Thanks.

-Scott

AdjuntoTamaño
Descargar snapshot00.jpg264.05 KB
Descargar snapshot01.jpg247.1 KB
publicaciones de 7 / 0 nuevos
Último envío
Para obtener más información sobre las optimizaciones del compilador, consulte el aviso sobre la optimización.

Is there someone from the Intel Perceptual Computing team who can comment on my question?

The bump in what should be planar data is smaller as I move the plane further from the camera, but it is still clearly noticeable. It occurs whether or not I use depth map smoothing.I am wondering if the camera has been properly calibrated.

Any information would be greatly appreciated.

Thanks.

-Scott

Hi Scott,

I did a couple of quick tests first with a flat board and then with me posing with the same normalised plane.  The flat board appears convex but not particularly wavy. Note I used PXCImage::COLOR_FORMAT_DEPTH for the data.

I hope this is of help.

David

 

Adjuntos: 

Thanks for doing the experiment David. At least now I know that I am not alone.

It is the convexity that concerns me. I should have used the word "bump" instead of "wave". It looks like there is a systematic error that is large when close to the camera. I would like to be able to correct for that error to start from more accurate 3D geometry.

-Scott

I suspect it is an overcorrection as the edges are further away from the centre of the lens than the central point of the plane but are registering as closer.

David

Hi, Scott,

I'.m with the "Wavy" or "Convex" situation (or so called issue) right now.

Since I am using the SenZ3D to roughly measure the volumn and depth of holes, this "plane measured as convex" problem is bothering me a lot. 

I thought it would be the distortion correction issue and trying to calibration instrinct parameters and distortion coefficients of the depth camera. If you found some way to overcome the problem, please contact me with wuzheming@gmail.com

Best,

Zheming

The first attached file illustrate my experment setup when using Senz3D to measure a flat plane, and the second shows the depth data result.

btw:

I using PXCImage::COLOR_FORMAT_VERTICES to restore the imdata.planes[0] and plot them in Matlab.

Thanks.

 

Adjuntos: 

AdjuntoTamaño
Descargar experiment setup.png4.68 KB
Descargar depth data.png238.66 KB

Deje un comentario

Por favor inicie sesión para agregar un comentario. ¿No es socio? Únase ya