ROS Depth Camera Integration


Use a Gazebo Depth Camera with ROS

Introduction

In this tutorial, you'll learn how to connect a Gazebo depth camera to ROS. The tutorial consists of 3 main steps:

  1. Create a Gazebo model that includes a ROS depth camera plugin
  2. Set up the depth camera in Gazebo
  3. View the depth camera's output in RViz.

This is a self-contained tutorial; it does not use the RRBot that is developed in other Gazebo ROS tutorials. It is designed to help you get up and running quickly using computer vision in ROS and Gazebo.

Prerequisites

You should install gazebo_ros_pkgs before doing this tutorial.

Create a Gazebo Model with a Depth Camera Plugin

Because Gazebo and ROS are separate projects that do not depend on each other, sensors from the gazebo_models repository (such as depth cameras) do not include ROS plugins by default. This means you have to make a custom camera based on those in the Gazebo model repository, and then add your own <plugin> tag to make the depth camera data publish point clouds and images to ROS topics.

You should choose a depth camera to use from those available in Gazebo. This tutorial will use the Microsoft Kinect, but the procedure should be the same for other depth cameras on the list.

First, acquire the depth camera and modify its name. We've packaged the Kinect sensor from gazebo_models repository for you, so all you have to do is download and unzip it.

Alternatively, you can follow the model contribution tutorial to make your own camera from scratch, or you can clone the gazebo_models repository and copy one of the sensors from there.

However you acquire it, copy the kinect folder into your ~/.gazebo/models directory. Then, change the name of your model to something meaningful, like kinect_ros. To change the model's name, you should update the folder name, the <name> stored in the .config file, and the model name in the model.sdf file.

Now you need to add the ROS plugin to publish depth camera information and output to ROS topics. A list of ROS plugins, with example code, can be found in the plugins tutorial.

In this tutorial, you'll be using the generic "Openni Kinect" plugin. You can (and should) use this plugin for other types of depth cameras besides the Kinect (it's an older plugin, and so it retains its old name).

Open the model.sdf file in your new model's directory. Add the following SDF markup inside the <sensor> tag, immediately after the closing </camera> tag:

        <plugin name="camera_plugin" filename="libgazebo_ros_openni_kinect.so">
          <baseline>0.2</baseline>
          <alwaysOn>true</alwaysOn>
          <!-- Keep this zero, update_rate in the parent <sensor> tag
            will control the frame rate. -->
          <updateRate>0.0</updateRate>
          <cameraName>camera_ir</cameraName>
          <imageTopicName>/camera/depth/image_raw</imageTopicName>
          <cameraInfoTopicName>/camera/depth/camera_info</cameraInfoTopicName>
          <depthImageTopicName>/camera/depth/image_raw</depthImageTopicName>
          <depthImageInfoTopicName>/camera/depth/camera_info</depthImageInfoTopicName>
          <pointCloudTopicName>/camera/depth/points</pointCloudTopicName>
          <frameName>camera_link</frameName>
          <pointCloudCutoff>0.05</pointCloudCutoff>
          <distortionK1>0</distortionK1>
          <distortionK2>0</distortionK2>
          <distortionK3>0</distortionK3>
          <distortionT1>0</distortionT1>
          <distortionT2>0</distortionT2>
          <CxPrime>0</CxPrime>
          <Cx>0</Cx>
          <Cy>0</Cy>
          <focalLength>0</focalLength>
          <hackBaseline>0</hackBaseline>
        </plugin>

As you can see, this plugin allows you a lot of fine-grained control over how the information is passed to ROS. A few points to note:

  • The updateRate parameter should be set to 0, which will cause the plugin to publish depth information as the same rate as the parent SDF sensor. If updateRate is not 0, it will do additional throttling on top of the parent sensor's update rate.
  • The topic names and frameName can be set to whatever you'd like, but the ones shown above match the default topics that are published by commonly used ROS packages, such as openni2_launch. Keeping the topic names the same will help make switching between real and simulated cameras easier.
  • The distortionX parameters should match those in the <distortion> tag of the parent camera. If there is no <distortion> tag, then use 0 for all the distortionX values.
  • pointCloudCutoff is the maximum distance for points. No points beyond this distance will be shown. This is an additional restriction to any clipping that has been set in the parent sensor.

Once you've renamed the model, added the above code to your .sdf file, and saved your changes, you should be ready to roll!

Set up the Depth Camera in Gazebo

Open Gazebo with ROS support enabled (e.g. roslaunch gazebo_ros empty_world.launch). Use the Insert panel to find your "Kinect ROS" model, and insert it into the world.

Important: You should also add some other objects to the scene, otherwise your camera might not have anything to see! Add some cubes, spheres, or anything else, and make sure they are located in the visible range of the camera, like in the screenshot below.

By default, the Kinect is not a static object in Gazebo. You may want to further edit your .sdf to add <static>true</static>, which will allow your camera to float in the air. This is probably much easier than recreating your entire sensing setup using physically correct models.

View Depth Camera Output in RViz

Now that the camera is in the Gazebo scene, it should be publishing images and point clouds to ROS topics. You can check the topics that are being published by running rostopic list in a new terminal. You should see the topics you specified in the SDF plugin code listed.

You can also run RViz (rosrun rviz rviz). First, set the RViz Fixed Frame in the left panel's Global Options section to match the value you set for <frameName> in the plugin XML code. Then, add a PointCloud2 and/or an Image display to RViz. For an Image, set the Image Topic to the value you used in the <imageTopicName> tag. For the PointCloud2, set the Topic to the name you used in the <depthImageTopicName> tag. See the screenshot below for an example that matches the values in the example sensor XML above:

After setting the correct topics and fixed frame, you should see something similar to the following from the PointCloud2:

An Image display will show a grayscale version of the depth camera results. If you have Gazebo 8 or newer, you can compare these RViz results to the depth image results available in the Gazebo Topic Visualizer.

Troubleshooting

Problem: rostopic list shows no camera topics.

Solution: Make sure you added the correct model in Gazebo. Make sure that the Gazebo simulation is running, not paused. Check the model.sdf file and ensure that the <plugin> tag is in the correct location in the file. Try running Gazebo in verbose mode (rosrun gazebo_ros gazebo --verbose) and see if there are any helpful warning or error messages that can help pinpoint the problem.

Problem: The ROS topics are listed, but I don't see anything in Rviz.

Solution: Make sure that there are objects for the camera to see in Gazebo. Make sure that you have an Image or PointCloud2 display added in RViz. Check that your Image or PointCloud2 displays are set to show the correct topic. Check that the Image or PointCloud2 displays are not disabled (checkbox). Ensure that your RViz Fixed Frame matches the frameName you specified in the <plugin> tag. Ensure that the sensor clipping parameters are not set up so that all points are being clipped.