recording a training bag

Setup your workspace

Please create a directory, where relatively large bag files will be recorded to

mkdir training_data
export WORK=`pwd`/training_data

Fiducial file

We need a discription of the fiducial markers that will be used for registering views of our object to capture.

If you are lucky to be using the willow lazy suzan you may find an appropriate description here

cd $WORK
wget http://vault.willowgarage.com/wgdata1/vol1/solutions_in_perception/training_data/vertical_sensor/fiducial.yml

Start the model_capture node

The model_capture program needs the fiducial description that we downloaded earlier. The default location will be in the working directory, however you may run model_capture with --help if you need to point it else where.

cd $WORK
rosrun tod_training model_capture camera:=/camera/rgb points:=/camera/rgb/points

The model_capture node requires the camera and points topics to be remapped. The above command will remap them to the openni topics.

Dynamic reconfigure (gui)

Now in a separate terminal run

rosrun dynamic_reconfigure reconfigure_gui

Verify that the openni_node is in sxga mode for best results.

Navigate to the model_capture node.

• out_dir is where bags will be saved
• object_name this should be a unique string that identifies your object (bags will be named after this)
• capture
• stop

Name your object and then press capture. Turn your turn table slowly, and when you are satisfied you may press stop. Don't rename your object unless you are stopped.

Debug

There are a few topics that are published by the capture node. Most interesting is the points_registered, which you may visualize in rviz.