Keypoint Recognition Using Two-Stage Randomized Trees

This paper proposes a high-precision, high-speed keypoint matching method using a two-stage Randomized Trees. The keypoint classification method uses the conventional Randomized Trees to enable highprecision, real-time keypoint matching. But the wide variety of view transformations for templates expressed by Randomized Trees make high-precision keypoint classification for all transformations difficult with a single Randomized Trees. To resolve this problem, proposed method classifies the template view transformations during the first stage. Then during the second stage, it classifies the keypoints using the Randomized Trees corresponding to each of the view transformations classified during the first stage. For images in which the viewpoint of the object is rotated by 70 degree, evaluation testing demonstrated that proposed method is 88.4% more precise than SIFT, and 63.4% more precise than the conventional Randomized Trees. We have also shown that the proposed method supports real-time keypoint matching at a speed of 12 fps.