Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Capture RGB from Kinnect with Openni and show with OpenCV

I need to capture the color RGB image from a Kinnect camera, but I want to show it in OpenCV as this is only a part of a bigger program. I know OpenCV has compatibility with OpenNI if you set the flag, but although I tried hard CMake couldn't find the path to OpenNI2 so I couldn't build OpenCV with OpenNI. Anyway I think it is good to know how to manually convert OpenNI frames to openCV frames, so I decided to follow this way.

For capturing the color frame in OpenNI I tried the following:

openni::Device device;  
openni::VideoStream  color;
openni::VideoFrameRef colorFrame;

rc = openni::OpenNI::initialize();
rc = device.open(openni::ANY_DEVICE);
rc = color.create(device, openni::SENSOR_COLOR);
rc = color.start();

color.readFrame(&colorFrame);
const openni::RGB888Pixel* imageBuffer = (const openni::RGB888Pixel*)colorFrame.getData();

But now I don't understand how to do the conversion to cv::Mat.

Does any anybody manged to do this?

like image 442
vgonisanz Avatar asked Apr 12 '13 08:04

vgonisanz


2 Answers

Ok, first you should separate initialization from the loop for reading frames, this way.

Initialization

openni::Device device;  
openni::VideoStream  color;
openni::VideoFrameRef colorFrame;

rc = openni::OpenNI::initialize();
rc = device.open(openni::ANY_DEVICE);
rc = color.create(device, openni::SENSOR_COLOR);
rc = color.start();

Mat frame;

Now comes the main loop for reading frames. You have done almost everything, the only thing left is to copy the buffer to the openCV Mat.

Loop for reading frames

while (true)
{
   color.readFrame(&colorFrame);
   const openni::RGB888Pixel* imageBuffer = (const openni::RGB888Pixel*)colorFrame.getData();

   frame.create(colorFrame.getHeight(), colorFrame.getWidth(), CV_8UC3);
   memcpy( frame.data, imageBuffer, 3*colorFrame.getHeight()*colorFrame.getWidth()*sizeof(uint8_t) );

   cv::cvtColor(frame,frame,CV_BGR2RGB); //this will put colors right
}
like image 73
Jav_Rock Avatar answered Sep 20 '22 17:09

Jav_Rock


Following up on @Jav_Rock's answer above, for IR and Depth the solution is similar, except for the following (in place of SENSOR_COLOR, CV_8UC3 and RGB888Pixel, respectively):

Depth

  • Sensor type: SENSOR_DEPTH
  • OpenCV type: CV_16UC1
  • VideoFrameRef data: DepthPixel

Note you'll probably want to set the PixelMode for the VideoStream's VideoFormat to PIXEL_FORMAT_DEPTH_100_UM otherwise your depth image will look very black.

IR

  • Sensor type: SENSOR_IR
  • OpenCV type: CV_16UC1
  • VideoFrameRef data: Grayscale16Pixel

Finally, note that neither depth nor IR need the cv::cvtColor call.

like image 42
Michael Gratton Avatar answered Sep 18 '22 17:09

Michael Gratton