Class COpenNI2_RGBD360

Defined in File COpenNI2_RGBD360.h

Inheritance Relationships

Base Types

public mrpt::hwdrivers::CGenericSensor (Class CGenericSensor)
public mrpt::hwdrivers::COpenNI2Generic (Class COpenNI2Generic)

Class Documentation

class COpenNI2_RGBD360 : public mrpt::hwdrivers::CGenericSensor, public mrpt::hwdrivers::COpenNI2Generic 

A class for grabbing RGBD images from several OpenNI2 sensors. This is used to obtain larger fields of view using a radial configuration of the sensors. The same options (resolution, fps, etc.) are used for every sensor.

Configuration and usage:

Data is returned as observations of type mrpt::obs::CObservationRGBD360. See those classes for documentation on their fields.

As with any other CGenericSensor class, the normal sequence of methods to be called is:

CGenericSensor::loadConfig() - Or calls to the individual setXXX() to configure the sensor parameters.
COpenNI2_RGBD360::initialize() - to start the communication with the sensor.
call COpenNI2_RGBD360::getNextObservation() for getting the data.

Calibration parameters

The reference system for both depth and RGB images provided by each individual OpenNI2 sensors are referred to the RGB Camera. The extrinsic parameters of each RGBD sensor are provided from a configuration file. This calibration was obtained using the method reported in [].

Coordinates convention

The origin of coordinates is the focal point of the RGB camera of the first indexed sensor, with the axes oriented as in the diagram shown in mrpt::obs::CObservation3DRangeScan. Notice in that picture that the RGB camera is assumed to have axes as usual in computer vision, which differ from those for the depth camera.

The X,Y,Z axes used to report the data from accelerometers coincide with those of the depth camera (e.g. the camera standing on a table would have an ACC_Z=-9.8m/s2).

Notice however that, for consistency with stereo cameras, when loading the calibration parameters from a configuration file, the left-to-right pose increment is expected as if both RGB & IR cameras had their +Z axes pointing forward, +X to the right, +Y downwards (just like it’s the standard in stereo cameras and in computer vision literature). In other words: the pose stored in this class uses a different axes convention for the depth camera than in a stereo camera, so when a pose L2R is loaded from a calibration file it’s actually converted like:

L2R(this class convention) = CPose3D(0,0,0,-90deg,0deg,-90deg) (+) L2R(in the config file)

Some general comments

Depth is grabbed in 10bit depth, and a range N it’s converted to meters as: range(m) = 0.1236 * tan(N/2842.5 + 1.1863)
This sensor can be also used from within rawlog-grabber to grab datasets within a robot with more sensors.
There is no built-in threading support, so if you use this class manually (not with-in rawlog-grabber), the ideal would be to create a thread and continuously request data from that thread (see mrpt::system::createThread ).
The intensity channel default to the RGB images, but it can be changed with setVideoChannel() to read the IR camera images (useful for calibrating).
This class relies on an embedded version of libfreenect (you do NOT need to install it in your system). Thanks guys for the great job!

Converting to 3D point cloud

You can convert the 3D observation into a 3D point cloud with this piece of code:

mrpt::obs::CObservationRGBD360      obs3D;
mrpt::maps::CColouredPointsMap       pntsMap;
pntsMap.colorScheme.scheme = CColouredPointsMap::cmFromIntensityImage;
pntsMap.loadFromRangeScan(obs3D, std::nullopt);

Then the point cloud mrpt::maps::CColouredPointsMap can be converted into an OpenGL object for rendering with mrpt::maps::CMetricMap::getAs3DObject() or alternatively with:

     mrpt::viz::CPointCloudColoured::Ptr gl_points =
*mrpt::viz::CPointCloudColoured::Create();
     gl_points->loadFromPointsMap(&pntsMap);

Platform-specific comments

For more details, refer to libfreenect documentation:

Linux: You’ll need root privileges to access Kinect. Or, install MRPT/scripts/51-kinect.rules in /etc/udev/rules.d/ to allow access to all users.
Windows:
- Since MRPT 0.9.4 you’ll only need to install libusb-win32: download and extract the latest libusb-win32-bin-x.x.x.x.zip
- To install the drivers, read this: http://openkinect.org/wiki/Getting_Started#Windows
MacOS: (write me!)

Format of parameters for loading from a .ini file

   PARAMETERS IN THE ".INI"-LIKE CONFIGURATION STRINGS:
  -------------------------------------------------------
    [supplied_section_name]
     sensorLabel     = RGBD360       // A text description
 
     device_number   = 0           // Device index to open (0:first Kinect,
 1:second Kinect,...)
 
     grab_image      = true        // Grab the RGB image channel?
*(Default=true)
     grab_depth      = true        // Grab the depth channel? (Default=true)
     grab_3D_points  = true        // Grab the 3D point cloud? (Default=true)
*If disabled, points can be generated later on.
 
     video_channel   = VIDEO_CHANNEL_RGB // Optional. Can be:
*VIDEO_CHANNEL_RGB (default) or VIDEO_CHANNEL_IR
 
     pose_x=0   // Camera position in the robot (meters)
     pose_y=0
     pose_z=0
     pose_yaw=0 // Angles in degrees
     pose_pitch=0
     pose_roll=0
 
 
     // Left/Depth camera
     [supplied_section_name_LEFT]
     rawlog-grabber-ignore = true // Instructs rawlog-grabber to ignore this
*section (it is not a separate device!)
 
     resolution = [640 488]
     cx         = 314.649173
     cy         = 240.160459
     fx         = 572.882768
     fy         = 542.739980
     dist       = [-4.747169e-03 -4.357976e-03 0.000000e+00 0.000000e+00
 0.000000e+00]    // The order is: [K1 K2 T1 T2 K3]
 
     // Right/RGB camera
     [supplied_section_name_RIGHT]
     rawlog-grabber-ignore = true // Instructs rawlog-grabber to ignore this
*section (it is not a separate device!)
 
     resolution = [640 480]
     cx         = 322.515987
     cy         = 259.055966
     fx         = 521.179233
     fy         = 493.033034
     dist       = [5.858325e-02 3.856792e-02 0.000000e+00 0.000000e+00
 0.000000e+00]    // The order is: [K1 K2 T1 T2 K3]
 
     // Relative pose of the right camera wrt to the left camera:
     // This assumes that both camera frames are such that +Z points
     // forwards, and +X and +Y to the right and downwards.
     // For the actual coordinates employed in 3D observations, see figure in
*mrpt::obs::CObservation3DRangeScan
     [supplied_section_name_LEFT2RIGHT_POSE]
     rawlog-grabber-ignore = true // Instructs rawlog-grabber to ignore this
*section (it is not a separate device!)
 
     pose_quaternion      = [0.025575 -0.000609 -0.001462 0.999987 0.002038
 0.004335 -0.001693]