OpenCV Perspective Transform (Markerless AR)

I’m currently translating a Markerless AR source code written in C++ to C#, but somehow my outputs are always off the point it should be. Below picture is my output:

Red circles are supposed to be the correct destination, but instead I got the black circles. And so I’m wondering if anyone here have any suggestions on what might causing this problem.

I’m not good at describing in words, so I try and describe it in this diagram:

Video instance: https://www.youtube.com/watch?v=3MaKHf3AIUU&feature=youtu.be
In this video:

• Left side image: Image shown are generated from Homography Matrix calculated in step 5 and step 6.
• Right side image: Image shown are generated from Homography Matrix calculated in step 8.

I uploaded my code onto GitHub: https://github.com/abc225313/MarkerLessDetector

1. Set marker and Train

   PatternDetector.cs → train(Pattern pattern)
PatternDetector.cs → buildPatternFromImage(Mat image, Pattern pattern)

2. Input image within main function

3. Feature Extraction

   PatternDetector.cs → bool extractFeatures (Mat image,ref KeyPoint[] keypoints, ref Mat descriptors)

4. Feature Matching

   PatternDetector.cs → void getMatches (Mat queryDescriptors, ref DMatch[] matches)

5. Find Homography

   PatternDetector.cs → within static bool refineMatchesWithHomography

6. Wrapping

   PatternDetector.cs → can be found within findPattern()

7. Calculating Homography

   PatternDetector.cs → calculated in function findPattern()`

8. PerspectiveTransform

   PatternDetector.cs → within static bool refineMatchesWithHomography