AN INLIER SELECTION METHOD BY USING 4-DIMENSINAL SUBSPACE(International Workshop on Advanced Image Technology 2006)

概要

論文の詳細を見る
In this paper, we propose an inlier selection method in feature points tracking of an image sequence. Kanade-Lucas-Tomasi's well-known feature points tracking method [1] (KLT method) may include tracking errors frequently in bad condition. Several research groups introduced outlier removing schemes into the KLT method by using 4-dimensional (4-D) subspace spanned by an orthogonal basis and RANSAC based method sampled 4 feature points out of many points. In our method, a few dozens of feature points are randomly sampled, and the 4-D subspace is spanned by a skew basis which has successfully identified feature point correspondences in our former research [2]. The skew basis provides a measure for deviation from the kernel of the 4-D subspace. Our numerical results show that the skew basis approximates 4-D subspace much better than the orthogonal one. The proposed method is effective as an inlier selection method accompanied by 3-D reconstruction scheme with repeatedly complementing additional feature points.
2006-01-03