Azure Kinect Crop Out Body Using Body Index Map

Question

I am using the Azure DK to read an MKV file where I am able to extract the RGBA and corresponding depth image for each frame. Thereafter I use the body tracker to locate and track people using the depth image. At the moment, I am able to generate a body index map and even convert it to the colour camera space as shown below.

k4a::image extract_body_index_map(tracker& object_tracker)
{
    k4a::image body_index_map = k4a::image();

    k4abt::frame body_frame = object_tracker.pop_result();

    if (body_frame != nullptr)
    {
        uint32_t bodies = body_frame.get_num_bodies();

        cout << "Bodies Found: " << bodies << "\n";

        for (uint32_t index = 0; index < bodies; ++index)
        {
            k4abt_body_t body = body_frame.get_body(index);

            cout << "\tPERSON: " << body.id << "\n";
            body_index_map = body_frame.get_body_index_map();
        }
    }
    else { cerr << "ERROR: Body Pop Out Timeout\n\n"; }

    return body_index_map;
}

k4a::image convert_body_index_map_to_colour(k4a::image& body_index_map, k4a::image& depth, k4a::image& colour, k4a_transformation_t transformation)
{
    k4a_image_t depth_image_in_colour_space = nullptr;

    k4a_result_t result = k4a_image_create(K4A_IMAGE_FORMAT_DEPTH16, 
        colour.get_width_pixels(), colour.get_height_pixels(), 
        colour.get_width_pixels() * (int)sizeof(uint16_t), &depth_image_in_colour_space);

    if (result == K4A_RESULT_FAILED)
    {
        cerr << "ERROR: Failed to create depth image in colour space\n";
        return nullptr;
    }

    k4a_image_t body_index_in_colour_space = nullptr;

    result = k4a_image_create(K4A_IMAGE_FORMAT_CUSTOM8,
        colour.get_width_pixels(), colour.get_height_pixels(),
        colour.get_width_pixels() * (int)sizeof(uint8_t), &body_index_in_colour_space);

    if (result == K4A_RESULT_FAILED)
    {
        cerr << "ERROR: Failed to create body index map in colour space\n";
        return nullptr;
    }


    result = k4a_transformation_depth_image_to_color_camera_custom(transformation, depth.handle(),
        body_index_map.handle(), depth_image_in_colour_space, body_index_in_colour_space,
        K4A_TRANSFORMATION_INTERPOLATION_TYPE_NEAREST, K4ABT_BODY_INDEX_MAP_BACKGROUND);

    if (result == K4A_RESULT_FAILED)
    {
        cerr << "ERROR: Failed to transform body index map to colour space\n";
        return nullptr;
    }

    return k4a::image(body_index_in_colour_space);
}

I am currently stuck on how to:

Use the generated body index map to crop/segment/draw out the colour image in order to extract the visual representation of the body. I am pretty much looking to replicate this; the portion with the RGBA images
Convert the cropped image into a 2D RGBA image

Any help will be greatly appreciated :)

eshirima · Accepted Answer · 2020-02-25T15:33:39.570

After messing around, I finally managed to resolve this. I switched over to OpenCV for image manipulation.

First, I converted body_index_in_colour_space to OpenCV Mat

Mat convertToMat(k4a::image kinect_image)
{
    uint8_t* pixel_data = kinect_image.get_buffer();

    if (pixel_data != nullptr) return Mat(kinect_image.get_height_pixels(), kinect_image.get_width_pixels(), CV_8UC4, (void*)pixel_data, Mat::AUTO_STEP);

    return Mat();
}

I converted the kinect BGRA image to OpenCV Mat as well in a similar manner by changing the flag from CV_8UC1 to CV_8UC4. Thereafter I simply drew the index map on top of the colour image.

Mat drawIndexMapOnColourImage(Mat indexMap, Mat colour)
{
    Mat result;
    colour.copyTo(result);

    //TODO: Not efficient, fix this
    for(size_t row = 0; row < indexMap.rows; ++row)
        for (size_t col = 0; col < indexMap.cols; ++col)
        {
            if ((int)indexMap.at<uchar>(row, col) == 255) continue; // ignore background

            result.at<Vec4b>(row, col) = 150;
        }

    return result;
}

Azure Kinect Crop Out Body Using Body Index Map

1 Answers1