This work aims to improve the matching correctness between scene images, which is the key issue in appearance-based localization tasks. Although existing approaches perform well when the test and map/model images are taken at near places, recognizing far-departed images is still challenging because of the deformed and missing features. In this paper, we propose to predict the type of feature deformation and select the robust features when the camera motion is limited to certain directions. The proposed approach can emphasize the importance of truly robust features in localization tasks. The experimental results shows that the proposed approaches outperforms the state-of-Art matching algorithms, especially in cases of matching distant images.
|Number of pages||6|
|Journal||IEEE International Conference on Automation Science and Engineering|
|State||Published - 1 Jan 2014|
|Event||2014 IEEE International Conference on Automation Science and Engineering, CASE 2014 - Taipei, Taiwan|
Duration: 18 Aug 2014 → 22 Aug 2014