foben's profile - activity

I want to write and read a video, not a single image. The code above just shows how I encode every frame. I see now how this is confusing, sorry about that.

Hey everyone,

I am trying to find a way to save and retrieve a video from a depth sensor (a kinect in my case) without losing any information.

The depthvideo I get is single channel with 16bit unsigned short per pixel (CV_16U). Since It is not possible to write such a video, I split the 16bits of every pixel and store two 8bit values in a standard 3 channel video with 8bits per pixel:

int rows = inputFrame.rows;
int cols = inputFrame.cols;
Mat outputFrame(rows, cols, CV_8UC3);
for (int r = 0; r < rows; r++)
{
    for(int c = 0; c < cols; c++)
    {
        unsigned short depthVal = inputFrame.at<unsigned short>(r, c);
        unsigned char val1 = depthVal >> 8;
        unsigned char val2 = depthVal & 0xff;

        outputFrame.at<Vec3b>(r, c)[0] = val1;
        outputFrame.at<Vec3b>(r, c)[1] = val2;
    }
}

To recover the frames, I combine the two values again and write them to a CV_16U image.

My problem is, that under linux, there seems to be no way to save an unencoded video, which of course destroys my 'encoding'. Under windows, when using -1 as the codec value of an ImageWriter, I can choose not to encode the video and I get Codec: 24 bits RGB (RV24). However, this does not work under Linux. When using 0 as codec value, the resulting image is Codec: Planar 4:2:0 YUV (I420). Recreating the depthvideo from these frames does not seem possible. I tried every COLOR_ constant to convert the image back to RGB, but none seems to recreate the correct image.

Is there a way on Linux to just write a video with raw rgb data? Or does anyone maybe have another idea of how to solve this problem? Any hints are greatly appreciated!

I have uploaded a sample image here: http://imageshack.us/f/547/5xyf.png/

As you can see, the cable is recorded multiple times in the depth image. What is maybe not so obvious is the fact that the duplication occurs only horizontally, not vertically.

I have built opencv with OpenNI support to grab (rgb and depth) images directly from the kinect. I am using the Kinect for windows version with the near mode feature.

I use the PrimeSense driver for the kinect.

Unfortunately, the depth images shows some errors: Some objects are duplicated in the depth image. The rgb image is fine.

I suspect that it may have to do with the near mode feature not being enabled. Is there a way to do this using opencv?

I use the following code to grab and display both the rgb and the depth image:

while(1)
{
    Mat depthImage;
    Mat bgrImage;

    capture.grab();
    capture.retrieve(depthImage, CV_CAP_OPENNI_DEPTH_MAP );
    capture.retrieve(bgrImage, CV_CAP_OPENNI_BGR_IMAGE );

    namedWindow("depth", 1);
    namedWindow("bgr", 1);

    imshow("depth", depthImage);
    imshow("bgr", bgrImage);

    if (waitKey(30) >= 0) break;
}

Any hints are greatly appreciated :)