부산대학교 도서관

This paper proposes a method for improving the egomotion determination by using latent semantic indexing for feature description and enhancing feature matching via an adaptive procedure. Input binocular stereo images are processed by the SURF detector, the descriptions are analyzed semantically with SVD and new feature descriptors are generated. Restrictions of matching pairs are established based on the epipolar geometry, which are then adaptively adjusted between successive image pairs. Four types of experiments are conducted and experiment results demonstrated the effectiveness of the proposed approach in providing smaller numbers of initial match candidates with higher probability of correct matches.