Evaluating the Influence of Feature Matching on the Performance of Visual Localization with Fisheye Images
M. Flores, D. Valiente, S. Cebollada, O. Reinoso, L. PayŠ
ICINCO 2021. 18th International Conference on Informatics in Control, Automation and Robotics  (Online streaming, 6-8 July, 2021)
Ed. SCITEPRESS - Science and Technology Publications, Lda.  ISBN:978-989-758-522-7  ISSN:2184-2809  - 434-441


Solving the localization problem is a crucial task in order to achieve autonomous navigation for a mobile robot. In this paper, the localization is solved using the Adaptive Probability-Oriented Feature Matching (APOFM) method, which produces robust matching data that permit obtaining the relative pose of the robot from a pair of images. The main characteristic of this method is that the environment is dynamically modelled by a 3D grid that estimates the probability of feature existence. The spatial probabilities obtained by this model are projected on the second image. These data are used to filter feature points in the second image by proximity to relevant areas in terms of probability. This approach improves the outlier rejection. This work aims to study the performance of this method using different types of local features to extract the visual information from the images provided by a fisheye camera. The results obtained with the APOFM method are evaluated and compared with the results obtained using a standard visual odometry process. The results determine that combining the APOFM method with ORB as local features provides the most efficient solution both to estimate relative orientation and translation, in contrast to SURF, KAZE and FAST feature detectors.