LINEMODTrainer

What Is This

A nodelet to train LINEMOD data from pointcloud and indices to mask the objects. This nodelet stores data of pointcloud and if you call ~start_training service, it will train the data and dump the templates into lmt file.

Subscribing Topic

• ~input (sensor_msgs/PointCloud2)

  This pointcloud should be able to be converted into pcl::PointXYZRGBA data.

• ~input/indices (pcl_msgs/PointIndices)

  Indices to mask object in ~input pointcloud.

• ~input/info (sensor_msgs/CameraInfo)

  Camera parameter to sample viewpoint.

Publishing Topic

• ~output/range_image (sensor_msgs/Image)

• ~output/colored_range_image (sensor_msgs/Image)

• ~output/sample_cloud (sensor_msgs/PointCloud2)

  Image and pointcloud generated by viewpoint sampling.

Advertising Servicies

• ~start_training (std_srvs/Empty)

  Start training and dump result into a file.

• ~clear (std_srvs/Empty)

  Clear stored data.

Parameters

• ~output_file (String, default: template)

  A file path to dump trained data.

  e.g.) If this parameter is set to /foo/bar, then /foo/bar.linemod, /foo/bar.pcd and /foo/bar_poses.yaml will be created.

• ~sample_viewpoint (Bool, default: True)

  Generate training data by samplingenerating viewpoint if this parameter is set to true.

• ~sample_viewpoint_angle_step (Double, default: 40.0)

• ~sample_viewpoint_angle_min (Double, default: -80.0)

• ~sample_viewpoint_angle_max (Double, default: 80.0)

• ~sample_viewpoint_radius_step (Double, default: 0.2)

• ~sample_viewpoint_radius_min (Double, default: 0.4)

• ~sample_viewpoint_radius_max (Double, default: 0.8)

  Viewpoint sampling parameters. Pose of model is sampled by golden ratio spatial technique.

• ~n_points (Int, default: 150)

  Number of viewpoint set.

Sample

roslaunch jsk_pcl_ros sample_linemod_trainer.launch