Region Analysis using point cloud data

Hello, I have the following issue.

I have a point cloud fetched from a Kinect. With this point cloud I would like to do a region analysis in order to find a specific object or to exclude objects. In order to do this I guess the best function would be “bwconncomp”.

First question: Is there a function to transform the point cloud into a 3d binary array in order to use “bwconncomp”?

Second question: Or do I have to program an algorithm on my own to get the 3D binary array?

Third question: Is there another/better approach to do a region analysis for 3D problems?

Thanks in advance

Answer 1

[Matt Cohen]

Hi Jürgen,

I understand that you are interested in performing region analysis on Kinect point cloud data.

The "bwconncomp" function and other Image Processing Toolbox functionality are probably not the best tools for performing this kind of analysis. There actually are 3-D Point Cloud Processing tools in MATLAB's Computer Vision System Toolbox.

If you have access to this toolbox, I would take a look at this documentation and see if any of the examples and functions seem appropriate for your data and desired task.

I hope this helps.

Matt

[Jürgen Buchner]

I do have these Toolboxes and I also use them, but the functions I found does not solve my problem. In my task i have to find a specific plate. Therefore, I use "pcfitplane" in order to fit planes in my Point Cloud data.

The problem is that I do not get information about the plates and therefore, I cannot differ between them. The only information are the inliers, but some of the inliers are not on my actual plate, but they are close to the plane and therefore counted as inliers. But I would need some information about min/max axis length, eccentricity, etc. to differ between the desired plate and a wall for example.

This is the reason why I would like to find objects regarding connectivity.

My steps would be finding inliers regarding "pcfitplane" analysis and then deleting (segmentation and filtering) the artefacts(inliers) which are not connected to the actual plate. After this I would look at features such as min/max axis length, etc. to differ wall and desired plate, for example.

Answer 2

[xuanfang he]

Hi Jürgen I got the same problems about segmenting the point clouds into connective clusters for a week without solution.did you solve it? would you like to help me?or share me the matlab code to study? Thanks in advance!