As known, for tracking objects in OpenCV we can use:
- FeatureDetector to find features
- DescriptorMatcher to match similarity between features of the desired object and features of current frame in video
- and then use findHomography to find the new position of the object
We can find the homography only if the majority of the keypoints has the same perspective transformation (homography). But if half of the points has the first transformation, and the other half has the second different transformation, how to find these two homography both? How to split an array of points to several arrays with different homography?
For example:
- To track the object of arbitrary shape I am using a rectangular reference image with the object, which contains an object and the background. We move the object, but the background stays in place - now we can not find the homography.
- Objects are separated (we are tracking moving motorcycle - motorcycle stopped and biker walked away - now we can not find the homography)
- Slightly changed the shape of the object (we are tracking the vehicle and it is not tracked after the opening of the trunk - now we can not find the homography)
How to find not one, but two or three main homography between the reference image of the object and the frame of the scene?
