ph4nt0m's profile - activity

Hello,

I'm using Ubuntu 10.04.3 and the ros-electric-openni-kinect package. When I try

roslaunch openni_camera openni_node.launch

I receive

[ INFO] [1320414269.335861019]: [/openni_node1] Number devices connected: 1
[ INFO] [1320414269.337386912]: [/openni_node1] 1. device on bus 001:08 is a Xbox NUI Camera (2ae) from Microsoft (45e) with serial id 'A003619160******'
[ INFO] [1320414269.343809795]: [/openni_node1] searching for device with index = 1
[openni_node1-2] process has died [pid 1734, exit code -4].

By searching the web, I found out that the reason for this is probably that my CPU does not support SSE3 instructions. As the only solution (apart from buying a new CPU) seems to be a manual source-based installation of the openni_kinect stack, I tried to follow the advice given here, but the CMakeLists.txt does not contain the line which is mentioned there.

Can anybody confirm that the pre-compiled package (installed by sudo apt-get install ros-electric-openni-kinect) requires SSE3 support? And, if so, what would be the easiest way to obtain a build that only makes use of SSE2 instructions?

Thanks in advance!

Hello,

how can I get the relative pose between two tf::Transforms? I know that tf::TransformListener can do this if you call lookupTransform with identical target- and source-frame, but with different target- and source-times, but how can I achieve this without any messages if I only have two tf::Transforms (each representing a different point in time) of an odometry-source for example?

Thanks in advance!

Thank you! The illustration already helps a bit, but why do you take the inverse at last? As far as I know, the transformations are applied from right to left if you transform a point x like this: T1 * T2 * x. So why isn't T12 = T2 * Inverse[T1] in your example?

Hello,

I would like to express the error between two rigid body transformations in terms of translation and rotation. What is the correct way to do this? I already know that you can determine the translational part (the norm of the 3x1 vector in the upper right corner) and rotational part (see angle theta here) of a single 4x4 matrix. But what kind of matrix should I use here?

Obviously, you have to consider some kind of "relative" transformation between the two original transformations. Most of the time, I see people using something like

Inverse[T1]*T2 or Inverse[T2]*T1,

which both lead to the same translation and angle. But what about the following?

T2*Inverse[T1] or T1*Inverse[T2]

Can somebody explain the difference between these two possibilities (maybe even geometrically)? Is there any reference for this kind of topic?

Thanks in advance!

Thanks! The build dir in pcl16 had to be deleted in order for the respective makefile target to be executed. Thank you for fixing this :) Just one further question: What would be the best way to upgrade to a new trunk? Just change the revision in the Makefile and recompile or to do a fresh install?

I did exactly what you said, but unfortunately when it comes to executing pcl16/Makefile, I get "CMake Error: The source directory "~/workspace/perception_pcl_fuerte_unstable/pcl16/build/pcl_trunk" does not appear to contain CMakeLists.txt". Apparently, the pcl trunk is not checked out at all. :(

I used "hg clone https://bitbucket.org/bhaskara/perception_pcl_fuerte_unstable", but I doesn't actually contain the pcl header and source files, so I think I have to get these from http://svn.pointclouds.org/pcl/trunk and merge them with perception_pcl_fuerte_unstable in some way!?

Thank you! How does this work? I cloned the git repository, but it doesn't contain the PCL trunk code, so I assume I have to merge them somehow. But how can this package change the c++ namespace to "pcl16" if it does not even contain the (modified) pcl header files?

Hello,

what is the easiest way to use the current experimental version of the PCL together with ROS Fuerte (which includes PCL 1.5 by default)? There are some features I would like to use which aren't available in version 1.5.

I found this directory but I don't know what these files are for, since there is no official PCL 1.5.2.

Thanks in advance!

But if the camera moves towards some features (so that they come nearer, their z values decrease), then the feature-transform would be one that moves every point in opposite z direction. But the position of the camera itself (seen from a global coord. sys.) would have moved forward in this example.

Hello,

I'm trying to learn more about the general concept of rgbdslam. Let's say I found a rigid body transformation (represented by a 4x4 matrix) between two consecutive frames (more precisely, between the feature sets of these, if I understand correctly) by using RANSAC, how can I use this to calculate the transformation that represents the change in pose between these two frames?

Thanks in advance!