Object recognition and pose estimation using KLT

This paper presents an object recognition method; feature X-D such as Kanade- Lucas-Tomasi Feature (KLT)-D and Speeded-Up Robust Features (SURF)-D. The main idea of the proposed algorithm is to use distance method to achieve rotation and position invariance. The anchor point, a center point of the target boundary region, is proposed in this paper as the basis of the pose estimation and it can be obtained by using KLT points. The proposed method shows more efficient performance for object recognition than SURF method in manipulation robot application. We also suggest a pose estimation method that requires no learning process and it is applicable for real time applications. We provide extensive experimental results to demonstrate performance in object recognition and pose estimation using constructed 44,486 images.