Sensor Data Collection

The package provides a ROS node for collecting data from one or more cameras/depth sensors.

ROS Node

Parameters

• device_name [type string, default "camera"]: the name of the device from which the data is acquired. It is the name of the directory where data are saved.
• save_directory [type string, mandatory, no default]: the root directory where all the data will be saved. Actually data are saved in save_directory/device_name/.
• continue [type bool, default false]: flag that set whether to add data to a previous acquisition.
• continue_from [type int, default 0]: if continue is true, continue_from is the first index that data will be saved with. As a special case, if continue_from is set to 0, the index is auto-computed.
• save_image [type bool, default false]: flag that set whether to save the RGB image provided by the sensor or not.
• save_image_camera_info, [type bool, default false]: flag that set whether to save the camera info data (relative to the RGB image) provided by the sensor or not.
• save_ir [type bool, default false]: flag that set whether to save the IR image provided by the sensor or not.
• save_ir_camera_info [type bool, default false]: flag that set whether to save the camera info data (relative to the IR image) provided by the sensor or not.
• save_depth_image [type bool, default false]: flag that set whether to save the depth image provided by the sensor or not.
• save_depth_camera_info [type bool, default false]: flag that set whether to save the camera info data (relative to the depth image) provided by the sensor or not.
• save_point_cloud [type bool, default false]: flag that set whether to save the point cloud provided by the sensor or not.
• depth_type [type string, default "uint16"]: the encoding format of the depth data. Two formats are supported:
  • "uint16": depth format of data provided by a Kinect.
  • "float32": depth format of data provided by a SwissRanger.

Subscribed Topics

• ~action [type: sensor_data_collection/Acquisition]: used by the node to trigger the acquisition. I.e. data are acquired as soon as a new message is published on this topic.
• ~image [type: sensor_msgs/Image]: topic where the RGB images are published.
• ~image_camera_info [type: sensor_msgs/CameraInfo]: topic where the camera info data (relative to the RGB images) are published.
• ~ir [type: sensor_msgs/Image]: topic where the IR images are published.
• ~image_camera_info [type: sensor_msgs/CameraInfo]: topic where the camera info data (relative to the IR images) are published.
• ~depth [type: sensor_msgs/Image]: topic where the depth images are published.
• ~image_camera_info [type: sensor_msgs/CameraInfo]: topic where the camera info data (relative to the depth images) are published.
• ~point_cloud [type: sensor_msgs/PointCloud2]: topic where the point clouds are published.

How-to

Create a Launch File

Create a file named my_device_data_collection_node.launch in sensor_data_collection/launch. Open it an fill it as following. Initiate the launcher.

<?xml version="1.0"?>
<launch>

Ask the user to provide the directory each time the node is launched.

  <arg name="save_directory" />

Let the user the opportunity to continue a previously interrupted acquisition.

  <arg name="continue"               default="false" />
  <arg name="continue_from"          default="0" />

Define the topics to save.

  <arg name="save_image"             default="true" />
  <arg name="save_image_camera_info" default="true" />
  <arg name="save_ir"                default="false" />
  <arg name="save_ir_camera_info"    default="false" />
  <arg name="save_depth_image"       default="true" />
  <arg name="save_depth_camera_info" default="true" />
  <arg name="save_point_cloud"       default="true" />

Define the name of the device.

  <arg name="device_name"            default="my_device" />

Create the node. Use the device_name argument to let multiple instances of the acquisition node to run at the same time.

  <node pkg="sensor_data_collection" type="data_collection_node" name="$(arg device_name)_data_collection"
        output="screen" required="true">

Use the previously defined arguments as parameters.

    <!-- Parameters -->
    <param name="device_name"             value="$(arg device_name)" />
    <param name="save_directory"          value="$(arg save_directory)" />
    
    <param name="continue"                value="$(arg continue)" />
    <param name="continue_from"           value="$(arg continue_from)" />
    
    <param name="save_image"              value="$(arg save_image)" />
    <param name="save_image_camera_info"  value="$(arg save_image_camera_info)" />
    <param name="save_ir"                 value="$(arg save_ir)" />
    <param name="save_ir_camera_info"     value="$(arg save_ir_camera_info)" />
    <param name="save_depth_image"        value="$(arg save_depth_image)" />
    <param name="save_depth_camera_info"  value="$(arg save_depth_camera_info)" />
    <param name="save_point_cloud"        value="$(arg save_point_cloud)" />

Set the depth type.

    <param name="depth_type"              value="uint16" />

First remap the ~action topic to a global /action one.

    <!-- Remapping topics -->
    <remap from="~action"            to="/action" />

Then remap all the other topics to the correct ones. We use the if="$(arg save_*)" to avoid subscription to unused topics.

    <remap from="~image"             to="/real_device/rgb/image_color"   if="$(arg save_image)" />
    <remap from="~image_camera_info" to="/real_device/rgb/camera_info"   if="$(arg save_image_camera_info)" />
    <remap from="~ir"                to="/real_device/ir/image_color"    if="$(arg save_ir)" />
    <remap from="~ir_camera_info"    to="/real_device/ir/camera_info"    if="$(arg save_ir_camera_info)" />
    <remap from="~depth"             to="/real_device/depth/image"       if="$(arg save_depth)" />
    <remap from="~depth_camera_info" to="/real_device/depth/camera_info" if="$(arg save_depth_camera_info)" />
    <remap from="~point_cloud"       to="/real_device/depth/points"      if="$(arg save_point_cloud)" />

Close the node and the launcher.

  </node>
</launch>

Run the Node

You can start the acuisition by running:

roslaunch sensor_data_collection my_device_data_collection_node.launch save_directory:=$HOME/.ros/sensor_data_collection/red_ball

All the directories are automatically created if not exisiting, otherwise they are checked if they are not empty.

You can continue a previously interrupted acquisition by running the same launcher:

roslaunch sensor_data_collection my_device_data_collection_node.launch save_directory:=$HOME/.ros/sensor_data_collection/red_ball continue:=true

The program will tell you the first available index: [my_device] First index is 3.

Trigger Acquisition

The acquisition is triggered by publishing an sensor_data_collection/Acquisition message on the defined topic:

rostopic pub /action sensor_data_collection/Acquisition "{}" -1

The Acquisition message has two fields:

• distance: distance in meters of the acquired object. [type: float]
• info: additional infos. [type: string]*

So, data can be acquired as following:

rostopic pub /action sensor_data_collection/Acquisition "{distance: 1.5, info: 'Red ball'}" -1

Get a Look at the Created Files

The node creates the following files:

• info.yaml: contains information about all the acquisitions performed.
• [image|ir|depth]_camera_info.yaml: contains the calibration parameters of the respective sensors.
• image_<n,4>.png: an image for each acquisition with padding zeros in the filename (e.g. n=5 -> image_0005.png). The image is compressed with a lossless encoding (png).
• ir_<n,4>.png: an image for each acquisition. The image is compressed with a lossless encoding (png).
• depth_<n,4>.png: an image for each acquisition. The image is compressed with a 16bit lossless encoding (png) .
• point_cloud_<n,4>.pcd: an point cloud for each acquisition.