Ask Your Question
2

Exporting RGBD-6D-SLAM scanned data to meshlab

asked 2011-03-16 06:17:53 -0500

duststorm gravatar image

updated 2014-01-28 17:09:22 -0500

ngrennan gravatar image

I'd like to know if there is a way to export the Point Cloud acquired with RGBD-6D-SLAM to a format usable in meshlab. That way I can triangulate it into a mesh and export it to whatever tool I want, eg Blender.

This is what I've already tried: Saving aqcuired data as .bag file rosbag record /rgbdslam/batch_clouds /tf

Extracting pointcloud data as .pcd file rosrun pcl_ros bag_to_pcd <bag_file_name> /rgbdslam/batch_clouds <destination_directory>

Here is where I'm stuck. If there is currently no way of doing this, I think I might try creating my own .pcd to .ply converter. I might be able to use the information at http://www.ros.org/wiki/pcl/Tutorials... for doing that.

edit retag flag offensive close merge delete

6 Answers

Sort by ยป oldest newest most voted
2

answered 2011-03-22 01:50:46 -0500

duststorm gravatar image

updated 2011-03-22 11:09:03 -0500

As to answering my own question:

I made a small .ply recorder that listens to a topic and writes out any received PointClouds to a .ply file.

Usage: pcl_plyrecord <topic name>

common usage for rgbdslam would be: ./pcl_plyrecord /rgbdslam/batch_clouds

Now you will typically start rgbdslam and record something with it (CTRL+P). Now, to get it to record a usable .ply file, "send" the data with the rgbdslam tool (CTRL+S). This will send the acquired point cloud to the previously mentioned topic and will make it record to a .ply file.

It will output to output.ply

CTRL+C to stop the program (try to stop it when no new data is coming in, or your output file might miss some)

WARNING: It will overwrite output.ply every time you start the program!

The program records xyzrgb type pointclouds, like the ones rgbdslam outputs. If you import the .ply in meshlab you will have a pointmap with color. Just what we need! :)

You can find the source package here. It contains the source and a buildscript and readme for quick compiling. (read the readme!!)

Or for those who just want to have a look at the source:

pcl_plyrecord.cpp

#include <ros/ros.h>
#include <pcl_ros/point_cloud.h>
#include <pcl/point_types.h>
#include <boost/foreach.hpp>
#include <stdio.h> 
#include "pcl_plyrecord.h"
#include <math.h>
//#include "rply.h"

typedef pcl::PointCloud<pcl::PointXYZRGB> PointCloud;

int count = 0;
FILE *file;
long int HEADER_COUNT_POS = 0;

/**
 * Add one point to output .ply file
 * Point will be omitted if it contains NaN coordinates
 **/
void addPoint(const float x, const float y, const float z, const float rgb)
{
    // Filter out points with NaN coordinates
    if(isnan(x) || isnan(y) || isnan(z) || isnan(rgb))
        return; // one of the values is NaN

    RGBValue color = RGBValue();
    color.float_value = rgb;
    unsigned char red = color.Red;
    unsigned char green = color.Green;
    unsigned char blue = color.Blue;

    fprintf(file, "%f %f %f %u %u %u\n", x, y, z, red, green, blue);
    count++;
}

/**
 * Write initial header of .ply file
 * Vertices count will be set to 0
 **/
void writeheader(void)
{
    fprintf (file, "ply\n");
    fprintf (file, "format ascii 1.0\n");
    fprintf (file, "comment Created with pcl_plyrecord\n");
    HEADER_COUNT_POS = ftell(file); // store location of vertices count
    fprintf (file, "element vertex %-10i\n", 0); // with extra padding for easy overwrite later on
    fprintf (file, "property float x\n");
    fprintf (file, "property float y\n");
    fprintf (file, "property float z\n");
    fprintf (file, "property uchar red\n");
    fprintf (file, "property uchar green\n");
    fprintf (file, "property uchar blue\n");
    fprintf (file, "end_header\n");
}

/**
 * Overwrite the vertices count in the header at the top of the .ply file
 * Because of extra padding to the number, larger numbers can always be written in-place
 * File pointer is set to the end of the file again afterwards
 **/
void correctHeader(void)
{
    fseek(file, HEADER_COUNT_POS, SEEK_SET);
    fprintf (file, "element vertex %-10i\n", count);
    fseek(file, 0, SEEK_END);
}

/**
 * Process a new incoming pointcloud
 * After adding all points of the cloud ...
(more)
edit flag offensive delete link more

Comments

As a sidenote: the output file could be made more compact if the program would output a binary instead of an ascii .ply file. Of course that makes it a little harder to read and parse afterwards. Meshlab supports both.
duststorm gravatar imageduststorm ( 2011-03-24 03:27:26 -0500 )edit
I've implemented your code, but I think there is a slight problem. You do not account for the transform anywhere, so the pointclouds are not mapped correctly in your combined ply file.
Shark gravatar imageShark ( 2011-04-14 01:43:23 -0500 )edit
Do you think so? I didn't notice that yet in my resulting files, but I must admit I haven't really been experimenting with it a lot. I do know however that when trying bag to pcd conversion there might be compaints about missing /transform data or such.
duststorm gravatar imageduststorm ( 2011-04-21 21:55:38 -0500 )edit
Maybe the fix would be to also record the /transform topic, extract the proper transformation information for an incoming pointcloud and multiply that pointcloud (or each point in the pointcloud) with the proper transformation matrix. I don't know ROS that well to be able to answer for sure.
duststorm gravatar imageduststorm ( 2011-04-21 21:59:53 -0500 )edit
If someone knows the correct workings of the transform and could explain, I could fix the code to account for transformations.
duststorm gravatar imageduststorm ( 2011-04-21 22:02:08 -0500 )edit
What you proposed above about also recording the transform is correct. However, this discussion is kind of moot now with the new version of rgbdslam, which let's to directly save the accumulated point cloud to .pcd or .ply
Shark gravatar imageShark ( 2011-05-05 06:35:56 -0500 )edit
1

answered 2011-04-15 00:26:34 -0500

Scott gravatar image

It is not exclusive to ROS, but Nicolas Burrus has a nice stand alone ogram you can use to easily convert captured Kinect 3D room scans into .ply files that can be imported into Meshlab. You could then potentially use that saved into to come back to ros for various applications.

Here is the link to his site:

http://nicolas.burrus.name/index.php/Research/KinectRgbDemoV4?from=Research.KinectRgbDemoV2

edit flag offensive delete link more

Comments

Thanks for the hint
raphael favier gravatar imageraphael favier ( 2011-04-15 03:34:37 -0500 )edit
1

answered 2011-03-16 10:00:58 -0500

raphael favier gravatar image

updated 2011-03-16 10:01:15 -0500

Hey Dust,

you could try editing the pcd file manually. In the end, it is just a text file with a header. Then, I assume there must be a way to import raw point clouds as text file with Meshlab.

For example, this is the beginning of sac_pane_test.pcd in pcl tests:

# .PCD v.5 - Point Cloud Data file format
FIELDS x y z w
SIZE 4 4 4 4
TYPE F F F F
WIDTH 3283
HEIGHT 1
POINTS 3283
DATA ascii
1.068 0.094721 -0.016231 0
1.061 -0.028772 0.060295 1
1.0623 -0.040527 0.061047 2
1.0662 -0.062634 0.068123 3

Raph

edit flag offensive delete link more
1

answered 2011-03-17 00:08:21 -0500

raphael favier gravatar image

updated 2011-03-17 00:28:41 -0500

Dust,

PCD creation

I don't really know the details of pcd export but can't you decide the format you want to use (binary or acsii)?

Did you create your pcd file by in the same way as in this tutorial? That way, you could save the pcd file directly inside RGBD-SLAM and not use a bag file. This should give you a better control on the pcd file you create.

Also have a look inside pcl/io, you will find the functions used to write pcd files from point clouds, maybe you can find how to force ascii mode.


Importing in Meshlab

I asked my colleague who actually used some point clouds I had generated with Kinect in Meshlab. All he did was:

  • Edit the pcd file (remove header and extra columns)
  • Change extension of the file to .asc

But as you said, you need a ascii format.

good luck :)

Raph

edit flag offensive delete link more
0

answered 2011-03-17 00:35:02 -0500

duststorm gravatar image

updated 2011-03-17 00:42:00 -0500

I created the pcd file using an existing bag to pcd converter in the packages.

I thought about this, and indeed, if I'm going to write my own .ply exporter anyway, I might as well skip all the unnecessary in-between steps. I could try creating a .pcd recorder, or even a .ply recorder, totally eliminating the need of a bag recorder.

Regarding your edit: that indeed makes it really simple. I had no idea that .pcd was already that close to a format that meshlab recognizes. I think I'll just mock up my own utility similar to what is shown in the tutorials to record a pcd as ascii (listen to a topic and write out an ascii pcd from the pointcloud data).

Thans a lot for your help.

edit flag offensive delete link more

Comments

yep, as you said, the pcd "recorder" is already available. I would also have a look inside the pcl stack of diamondback. I can see executables like "convert_pcd_ascii_binary" or "concatenate_points_pcd" and "alu_pointcloud_to_pcd" which I am sure you can inspire from to create your ply exporter. Could it be possible for you to post your ply exporter once done? I guess we are quite a lot that would benefit from such a tool. Good luck!
raphael favier gravatar imageraphael favier ( 2011-03-17 00:46:31 -0500 )edit
Yeah, sure. I was already planning to post my source. I'll post it in this topic when I have something.
duststorm gravatar imageduststorm ( 2011-03-17 01:46:28 -0500 )edit
0

answered 2011-03-16 22:03:15 -0500

duststorm gravatar image

Thanks Raph,

I have been looking at my pcd file too. Apart from the fact that my point cloud has some different fields (it has an RGB value for each point and no 'w'), the data itself is binary and by consequence is a little bit harder to parse.

These are the contents of my pcd:

# .PCD v.7 - Point Cloud Data file format
FIELDS x y z rgb
SIZE 4 4 4 4
TYPE F F F F
COUNT 1 1 1 1
WIDTH 8294400
HEIGHT 1
VIEWPOINT 0 0 0 1 0 0 0
POINTS 8294400
DATA binary
...

I'll probably have to make a conversion programme if such a thing does not exist yet.

edit flag offensive delete link more

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

5 followers

Stats

Asked: 2011-03-16 06:17:53 -0500

Seen: 4,118 times

Last updated: Apr 15 '11