QuadrangleInPicture

Quick introduction

Shows how to embed a quadrangle video into a picture using OpenCV C++.

This is the output.

Description

You have an image and want to embed on top of the image another image. **Note that when I say image, it can also be video as video is a series of images. If the image you want to embed is a square or rectangle, there are other simpler methods. This method is overkill for square or rectangle. But if the image you want to embed is a quadrangle(four sided polygon) but not square or rectangle, this method suits you. We will use homography matrix and warped perspective.

Your inputs:

1. background image
2. video to be embed
3. The coordinate of the embedded target.

Steps

1. Setup the background image and video.

    // setup background image
    String backgroundFilename("background.jpg"); 
    Mat background = imread(backgroundFilename, CV_LOAD_IMAGE_COLOR);
   
    // setup video capture
    VideoCapture cap("incredibles-2-trailer-1_h480p.mov");
    if (!cap.isOpened())
    {
    	cout << "Error opening";
    	return -1;
    }
   
    // the video we want to put in background
    Mat video;
    cap >> video;
   

2. Define the video region (we use whole video resolution) and its corresponding region in the background image.

    // video source
    vector<Point> left_image;      
    left_image.push_back(Point(0, 0));
    left_image.push_back(Point(0, video.rows));
    left_image.push_back(Point(video.cols, video.rows));
    left_image.push_back(Point(video.cols, 0));
   
    // corresponding video coordinate destination
    vector<Point> right_image;
    right_image.push_back(Point(100, 100));
    right_image.push_back(Point(200, 500));
    right_image.push_back(Point(900, 650));
    right_image.push_back(Point(700, 200));
   

3. Calculate homography matrix.

    // once we get 4 corresponding points in both images calculate homography matrix
    Mat H = findHomography(left_image, right_image, 0);
   

4. Warped perspective.

    // warped perspective
    Mat warpedImage;
    warpPerspective(video, warpedImage, H, background.size());
   

5. Paint the region in the background image where we want to paste the video to black.

    // create a mask at background
    int num = 4;
    const Point *pt4 = &(right_image[0]);
    fillPoly(background, &pt4, &num, 1, Scalar(0, 0, 0), 8);
   

6. Sum background image with warped perspective image. (Any colored pixel summed with black pixel will results to the colored pixel).

Compile notes

The project is setup based on OpenCV official tutorial (property sheets setup). Check Visual Studio property manager and assign your configured property sheets.