A generalized implementation of RANSAC (Random Sample Consensus),
In order to retrieve homography image transformations from sets of points without descriptive local feature vectors to allow for point pairing, we introduce two logical criteria. We propose a robust criterion that rejects implausible point selection before each iteration of RANSAC, based on the type of the quadrilaterals formed by random point pair selection (convex or concave and (non)-self-intersecting). To cover some practical applications we allow the points to be (optionally) labelled in two classes. Also, a similar post-hoc criterion rejects implausible homography transformations is included at the end of each iteration. Both criteria are calculating the value Q, based on the below equations, for each shaped quadrilateral
where
with
An example of different Q values, is depicted on the below figure.
Let image A and image B be a pair of images that a homography transformation can be applied. Also, let the ordered 4-points
To gain an insight of the expected reduction in the number of iterations of the proposed
The proposed methodology is compared with variants of RANSAC algorithm (like VSAC, MAGSAC, Graph-Cut or DEGENSAC), utilizing feature extraction techniques like SIFT, ORB or more recent deep learning techniques like SuperPoint and SuperGlue. The dataset that was used to extract and validate our theoritical proposal, contains football images acquired from 12 different cameras (capturing images at the same time) with different camera positions, viewing vector and zoom factor. An example of the resulted image, as well as the images that used are displayed below.
The reference frame for the specific timestamp is the below
And the result, if we combine the reference frame with the transformed images
If you use the above software or part of it in your work, please cite it using the below
Nousias, G., Delibasis, K., & Maglogiannis, I. (2023). H-RANSAC, an algorithmic variant for Homography image transform from featureless point sets: application to video-based football analytics. arXiv [Cs.CV]. Retrieved from http://arxiv.org/abs/2310.04912