problem in SACMODEL_NORMAL_PARALLEL_PLANE

[ClaireXie]

I tried to detect two perpendicular planes in which the normal of the first plane is parallel to the second plane. So I set the normal of the first plane as the axis in SACMODEL_NORMAL_PARALLEL_PLANE and use SACSegmentationFromNormals. However, after it, I find the two detected planes are almost parallel to each other, instead of perpendicular. When I dig into the code, in sample_consensus/impl/sac_model_normal_parallel_plane.hpp, in function isModelValid(), you are checking

if (fabs (axis_.dot (coeff)) < cos_angle_)
      return  (false);

which, I feel strange.
I thought it should be (consistent with 'sac_model_parallel_plane'):

if (fabs (axis.dot (coeff)) > sin_angle_)
      return  (false);

[VictorLamoine]

I tried to detect two perpendicular planes in which the normal of the first plane is parallel to the second plane

If the two planes are perpendicular; they can't be parralel. The inverse is also true.

I find the two detected planes are almost parallel to each other, instead of perpendicular

This seems pretty normal when you use a SACMODEL_PARALLEL_PLANE

Two planes cannot be both parallel and perpendicular.

[ClaireXie]

I tried to detect two perpendicular planes in which the normal of the first plane is parallel to the second plane

If the two planes are perpendicular; they can't be parralel. The inverse is also true.

SACMODEL_NORMAL_PARALLEL_PLANE should constrain the plane parallel to a specified AXIS. If one plane is parallel to the other plane's normal, it means that these two planes are perpendicular. So that's why I use SACMODEL_NORMAL_PARALLEL_PLANE to detect two perpendicular planes.

[VictorLamoine]

If one plane is parallel to the other plane's normal, it means that these two planes are perpendicular

Or in a simpler way; the dot product between the 2 planes normals is null.
Why don't you use SACMODEL_PERPENDICULAR_PLANE ?

[ClaireXie]

Or in a simpler way; the dot product between the 2 planes normals is null.
Why don't you use SACMODEL_PERPENDICULAR_PLANE ?

Sorry for the confusion. I am trying to fit two perpendicular planes. In other words, I am trying to estimate the two planes' coefficients.

[VictorLamoine]

This is the point of SACMODEL_PERPENDICULAR_PLANE.

seg.setModelType (pcl::SACMODEL_PERPENDICULAR_PLANE);
seg.setMethodType (pcl::SAC_RANSAC);
seg.setMaxIterations(1000);
seg.setDistanceThreshold (0.05);
seg.setAxis(Eigen::Vector3f(0,0,1));
seg.setEpsAngle (0);

In setAxis, put the normal of your first plane. (the first 3 coefficients in the plane coefficients).

[ClaireXie]

In setAxis, put the normal of your first plane. (the first 3 coefficients in the plane coefficients).

I did that exactly as you mentioned above. But the problem that I found in the PCL implemntation has been stated in my first post. (as the a result, the fitted two planes are parallel instead of being perpendicular.)

[VictorLamoine]

I took a look at whats going on here; first I don't get why there is both

SACMODEL_PERPENDICULAR_PLANE
SACMODEL_NORMAL_PARALLEL_PLANE

I might be wrong but to me being parallel to the normal of a plane is strictly equivalent to being perpendicular to a plane.

Second thing that bugs me: in the isModelValid function there are two possibilities:
if (eps_angle_ > 0.0)
Then the dot product will be computed to check wether we fit the axis or nor (with sin_angle_ as a threshold)

If eps_angle_ < 0.0 then the model returned is always valid, it is not checked whether is fits the axis or not ?! (This means that if you don't specify an eps_angle_ the model will always be valid whatever the plane is)

Have you had any success with SACMODEL_PARALLEL_PLANE ? I can't get it working properly

[VictorLamoine]

Calling setAxis () alone is not enough, the axis is ignored until the user sets the epsilon angle:
A first fix would be to add setEpsAngle (std::min) when calling setAxis (), this would enable the axis check to the lowest possible tolerance.

Now here is a question: When you are looking for a parallel plane giving the axis, do you want:

• The normal of the first plane to be parallel with the axis
• The axis to fit into the plane (in that case the plane and the axis are parallel)

I don't know the answer to that question.
http://www.pcl-users.org/Ransac-Planes-tp2085912p2086759.html

[mvieth]

This thread is rather old so maybe no one is interested any more, but for what it is worth:
SACMODEL_NORMAL_PARALLEL_PLANE is like SACMODEL_PERPENDICULAR_PLANE, but the points are only counted as inliers if they are close to the plane and their normals are close to the normal of the estimated plane. See also SACMODEL_NORMAL_PLANE