CloudCompare forum

Hi

I would like to know more about how can Cloudcompare do the histogram. As mentioned at page 108 in user manual (2.6.1), Hausdorff distance is applied. If I have two point cloud sets, I think I only can find one Hausdorff distance value for this two cloud sets. If this is correct (or wrong), how can cloudcompare calculate how many points are located at specific C2C absolute distances?

Thank you.

In fact the 'Haussdorf' distance here is the set of distances computed for every points. We don't take the maximum distance out of it as one would do with the real Haussdorf distance (even though you can get this value easily afterwards).

Thank you so much for the reply.

I have three more questions described as follows,

1. I have read your paper recently,and you mention the equation as can been seen in Figure 1. However, it seems that the equation is a little bit different from the definition of Haussdorf distance. The equation you mentioned in the paper only calculated the shortest distance between the compared cloud and the reference cloud. I would like to confirm if the CloudCompare use the same concept?

2. I intent to offset two cubes to analyze the C2C distance as shown in Figure 2. Each edge of the two cubes is 2 cm. (so I offset the compared cube 2cm to the right side of the reference cube). As you can see the result, it seems that the C2C distance only compare the compared cube with only right surface of the reference cube. Is there any way that CloudCompare can automatically detect the real surface (or points) that user wants to compare with?

3. When doing the "subsample", I found out every time the subsample points number is different (I always use the default number to subsample the object). Is this a bug?


Thank you for solving my puzzles.

1) This is not the shortest distance from the compared cloud, but only for one point 'p'. You would then have to take the maximum value over all the points to compute the actual Hausdorff distance. As I explained we do compute the shortest distance for all points, but we don't actually display or use the maximum value (even though you can get it directly when looking at the distances scalar field properties, or its histogram).

2) The C2C distances indeed computes for each point the distance to the nearest point in the other cloud (see above). Therefore the nearest point is always on the nearest side. To get a smarter behavior you should look at M3C2 (http://www.cloudcompare.org/doc/wiki/in ... 2_(plugin)).

3) Are you speaking about the point cloud 'subsample' tool? Or the mesh 'sample' tool?
- the first one gives me a constant result when applying it to the same cloud several times.
- the second (Mesh > Subsample) tries to generate the specified number of points... But as it needs to sample an integer number of points per triangle and this number depends on the triangle surface, there is some statistical drawing at stake to determine if we should sample one less point or one more point. In the end the total sum may not be the exact required number. And it changes each time because of the random draw.