pcopen - once again

[sanostol]

Hi, I'm trying to get this working for some time right now, but without succss.

I use the pcopen function, but I want to use the not very well documented "ndot" argument. My goal is to get every point to use a look direction and cone, so that every point has a limited angle of view. only points of the point cloud that are inside this cone are considered, but I don't get it to work.

my function looks like this:

int handle = pcopen(pcFile, pChannel,P, nChannel, N, maxdistance, maxpoints,"ndot", cos(radians(coneangle)) );

how can I achieve this effect?

martin

[pclaes]

You have to do that manually, the pcopen does not support it.

It will make its' selection on all the points facing in the same-ish direction, rather than select the points in the "field of view".

So to do it manually:

*)open a normal pointcloud, big enough to encompass all the points in the sphere of influence (both in radius as well as in amount of points).

*)setup your pointcloud while loop and pcimport the P.

*) calculate the direction vector from current point A to imported point B =>

B-A

(you do not want to normalize this yet as you will want to use the length of this direction vector (effectively the distance to the other point) as a weight - similar to what the pcfilter does )

*) compare the original vector from A with the direction vector to B =>

dot( normalize(A) , normalize( ) 

*) the result of the dot product will be ranging from -1 to 1, a result of 1 means the point is right in front of A, 0 means it is perpendicular (at 90degree angle), -1 is behind A. You can use ramps or fit functions to remap the influence based on this.

*) To make it a bit more user friendly you may want to use degrees instead to specify that angle of influence. From the

dot(A, = |A|.|B|.cos(alpha)

follows (if you are using normalized vector which you should)

alpha = acos(dot(A,)

. Alpha will be returned in radians.

*) Then with an if statement you can reject:

if( alpha > degreesToRadians(60))

Overall it's still pretty fast as you are still using pointclouds to narrow the huge cloud to the sphere of influence.

Edited June 9, 2010 by pclaes

[sanostol]

thanks peter, great!