Capturing stereo video

GIFT-Grab allows for capturing stereo video streams from frame grabbers supporting this feature. The example below demonstrates how stereo frames can be acquired and saved to individual image files. Running this example requires GIFT-Grab built/installed with support for Blackmagic DeckLink 4K Extreme 12G and NumPy. The example uses OpenCV to save video frames to disk.

The full source code of the example is below. Please follow the comments and the flow of code. This example is also available on GitHub:

#!/usr/bin/env python2
# -*- coding: utf-8 -*-

"""
Example demonstrating how stereo video frames can be captured
using a frame grabber card that supports this feature.
"""

import time
import cv2
import numpy as np
from pygiftgrab import (IObserver, VideoSourceFactory,
                        ColourSpace, Device, VideoFrame)


class StereoFrameSaver(IObserver):
    """
    Simple class that demonstrates how mono and stereo frames,
    and their respective parameters can be queried and the actual
    frame data can be saved using the GIFT-Grab stereo API.
    """

    def __init__(self):
        super(StereoFrameSaver, self).__init__()
        self.current = 0

    def update(self, frame):
        self.current += 1

        # 4 is the number of variations of stereo/mono
        # calls to the data method, using it here as well to
        # avoid flooding the user's terminal
        if self.current <= 4:
            # display number of stereo frames, should be 2
            # for this device
            print(
                'Got {} stereo frames'.format(
                    frame.stereo_count()
                )
            )
            # display length of data of each stereo frame,
            # each stereo frame should consist of same number
            # of bytes for this device
            print(
                'Stereo data length (bytes):\n'
                '\tdata_length(): {}\n'
                '\tdata_length(0): {}\n'
                '\tdata_length(1): {}\n'.format(
                    frame.data_length(), frame.data_length(0),
                    frame.data_length(1)
                )
            )

        frame_shape = (frame.rows(), frame.cols(), 4)

        # the slicing below, i.e. [:, :, :3], is due to OpenCV's
        # imwrite expecting BGR data, so we strip out the alpha
        # channel of each frame when saving it

        if self.current == 1:
            # all three calls below save the same frame,
            # that is the first of the two stereo frames
            cv2.imwrite(
                'mono-frame.data.png',
                np.reshape(frame.data(), frame_shape)[:, :, :3]
            )
            cv2.imwrite(
                'mono-frame.data-False.png',
                np.reshape(frame.data(False), frame_shape)[:, :, :3]
            )
            cv2.imwrite(
                'mono-frame.data-False-0.png',
                np.reshape(frame.data(False, 0), frame_shape)[:, :, :3]
            )

        elif self.current == 2:
            # the two calls below save the two stereo frames,
            # however the data needs to be reshaped, as the
            # call to the data method yields a flat NumPy array
            cv2.imwrite(
                'stereo-frame.data-False-0.png',
                np.reshape(frame.data(False, 0), frame_shape)[:, :, :3]
            )
            cv2.imwrite(
                'stereo-frame.data-False-1.png',
                np.reshape(frame.data(False, 1), frame_shape)[:, :, :3]
            )

        elif self.current == 3:
            # the two calls below save the two stereo frames,
            # without the need for reshaping the data, as the
            # call to the data method already yields a
            # structured NumPy array
            cv2.imwrite(
                'mono-frame.data-True.png',
                frame.data(True)[:, :, :3]
            )
            cv2.imwrite(
                'mono-frame.data-True-0.png',
                frame.data(True, 0)[:, :, :3]
            )

        elif self.current == 4:
            # the two calls below save the two stereo frames,
            # without the need for reshaping the data, as the
            # call to the data method already yields a
            # structured NumPy array
            cv2.imwrite(
                'stereo-frame.data-True-0.png',
                frame.data(True, 0)[:, :, :3]
            )
            cv2.imwrite(
                'stereo-frame.data-True-1.png',
                frame.data(True, 1)[:, :, :3]
            )


if __name__ == '__main__':
    sfac = VideoSourceFactory.get_instance()
    source = sfac.get_device(
        Device.DeckLink4KExtreme12G, ColourSpace.BGRA
    )

    saver = StereoFrameSaver()

    source.attach(saver)

    time.sleep(2)  # operate pipeline for 2 sec

    source.detach(saver)