Combing Line and Point Correspondences to Calculate Homographies with Applications in Hockey Rink Registration

The goal of this project is to explore the use of a combination of line and point correspondences to improve the computation of homography transformations. The rink registration aspect of the UBC Hockey Tracking System currently uses only keypoint matches between frames leaving the line information on the rink unutilized. The hypothesis is that use of line information will help make the homography calculation more robust, especially in a closeup situation where not many keypoints are visible. The current system uses the normalized DLT (Direct Linear Transform) algorithm to calculate homographies from point correspondences that are acquired using the Kanade-Lucas-Tomasi (KLT) tracker. A RANSAC algorithm is used to prune outlier correspondences from the calculation. Both the DLT and RANSAC algorithms are extended to use line correspondences as well as points. Experimental results are presented for a simple chessboard image. Rink registration for the hockey system is not yet improved due to a weakness in the DLT algorithm regarding normalization.