OASIS Kinect2 for ROS 2

This repo is a ROS 2 fork of the Institute for Artificial Intelligence - University of Bremen Kinect2 driver.

It is used by the OASIS smarthome OS.

See below for the status of this port.

Dependencies

All OASIS dependencies, including this one, are managed by the oasis_tooling package.

The specific versions are defined in the file depends.repos. The dependencies for Kinect 2 functionality are:

repositories:
  # Requires: libturbojpeg0-dev libusb-1.0-0-dev ocl-icd-opencl-dev
  depends/libfreenect2:
    type: git
    url: https://github.com/OpenKinect/libfreenect2.git
    version: master
  # Requires: default-jdk libudev-dev libusb-1.0-0-dev
  depends/OpenNI2:
    type: git
    url: https://github.com/structureio/OpenNI2.git
    version: master
  ros-perception/image_pipeline:
    type: git
    url: https://github.com/ros-perception/image_pipeline.git
    version: rolling
  ros-perception/image_transport_plugins:
    type: git
    url: https://github.com/ros-perception/image_transport_plugins.git
    version: rolling

Installation, GPU acceleration and Citations

The original README can be found at: https://github.com/code-iai/iai_kinect2

What is working?

Currently, four topics are working:

/hd/image_color:

header:
  frame_id: kinect2_rgb_optical_frame
height: 1080
width: 1920
encoding: bgr8
is_bigendian: 0

/qhd/image_color:

header:
  frame_id: kinect2_rgb_optical_frame
height: 540
width: 960
encoding: bgr8
is_bigendian: 0

/sd/image_color:

header:
  frame_id: kinect2_rgb_optical_frame
height: 360
width: 640
encoding: bgr8
is_bigendian: 0

/sd/image_color_rect:

header:
  frame_id: kinect2_ir_optical_frame
height: 424
width: 512
encoding: bgr8
is_bigendian: 0

Here is a combined visualization of all topics. Most aren't working. HD and QHD color rectified are a duplication of the color image. Subscribing to HD and QHD depth rectified crashes the driver.

Launch file

Due to scalability issues in the existing code, and being written before modern APIs, downscaling to SD happens in a new node. It downscales from QHD (quarter HD) to SD (in this case 640x360).

If you want the SD color topic, you can either launch the downscaler as a node, or combine it with the bridge as components in a single process to take advantage of zero-copy.

def generate_launch_description() -> LaunchDescription:
    ld = LaunchDescription()

    # Example using separate nodes
    kinect2_bridge_node = Node(
        namespace=ROS_NAMESPACE,
        package="kinect2_bridge",
        executable="kinect2_bridge",
        name="kinect2_bridge",
        output="screen",
    )
    ld.add_action(kinect2_bridge_node)

    kinect2_downscaler_node = Node(
        namespace=ROS_NAMESPACE,
        package="kinect2_bridge",
        executable="kinect2_downscaler",
        name="kinect2_downscaler",
        output="screen",
    )
    ld.add_action(kinect2_downscaler_node)

    # Example using a container
    kinect_container = ComposableNodeContainer(
        namespace=ROS_NAMESPACE,
        name="kinect2_container",
        package="rclcpp_components",
        executable="component_container",
        output="screen",
        composable_node_descriptions=[
            ComposableNode(
                package="kinect2_bridge",
                plugin="kinect2_bridge::Kinect2BridgeComponent",
                name="kinect2_bridge",
            ),
            ComposableNode(
                package="kinect2_bridge",
                plugin="kinect2_bridge::Kinect2DownscalerComponent",
                name="kinect2_downscaler",
            ),
        ],
    )
    ld.add_action(kinect_container)

NOTE: Due to the use of old APIs, the kinect2_bridge node doesn't work as a component. Modernization is in progress.