OctoMap and PCL Question

Hi,

I came across a paper, "Hierarchies of Octrees for Efficient 3D Mapping"... it uses octrees to map the environment. I'm interested in this particular subject, and i'm exploring pcl::octrees in a application i'm developing. Is pcl::octrees based on Octomap library or is it another completly different type of implementation?

The reason i'm asking this, is because on my implementation i've converted an *.stl file in pointcloud, passed it through an octree with a rather small leaf size (0.02). The objective is to get a voxelized object. After i get the computed OccupiedVoxelCenters (implemented in pcl/octree/octree_pointcloud.hpp) i managed to see them on the visualizer, mainly through VTK. My problem is, since i don't have enough points in the pointcloud (for the intended lead size) i get a lot of holes in the voxelized object, voxel centers where not computed because they didn't had points inside the voxel. I'm currently trying to find a way to add the voxelcenters that were missed. (strangly all holes were closed only with a leaf size of 0.6)
Is Octomap better on this kind of perfomance?

Sorry to refer to Ocmap here, but on the paper the name of Radu Rusu came up.

Cheers

Octrees are general purpose data containers/structures which have certain properties (eight octant subdivision in this
case). As trees, they can be used for many applications, some involving visualization, compression, data reduction,
change detection, and so on.

Octomap is a code library that implements a 3D occupancy grid using octrees for mapping. This means that an octree
structure is used for a particular application (mapping) in this case, and thus most likely tailored with respect to the
specifics of that application.

PCL is a collection of libraries (think as "Boost 3D" being our final goal). One of the PCL libraries is called Octree
(libpcl_octree), and has a collection of different types of octrees. These octrees can be used for many different
applications like compression, change detection, mapping, etc.

More recently, the PCL community has been working on an out-of-core octree implementation (libpcl_outofcore) for large
scale mapping, i.e. the datasets/maps do not fit in memory and you have to store parts of them on disk. The goal is to
easily stream and use datasets of billions of points without any problems. We are collaborating with a lot of commercial
companies on this open source effort. See http://www.pointclouds.org/blog for more details.

You should figure out exactly what your application needs, then dig into our documentation to see which _type of octree_
is most suitable for you. There is not a single library or type of octree that can fulfill the requirements of many
different applications, so you need to pick what works best.

In terms of efficiency, PCL would be happy to optimize any piece of code that our users might find inefficient.

Cheers,
Radu.
--
http://pointclouds.org

On 03/22/2012 03:27 PM, Zhao wrote: _______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users