Evaluation of Image Detection and Description Algorithms for Application in Monocular SLAM

Evaluation of Image Detection and Description Algorithms for Application in Monocular SLAM

Claudio Urrea (Universidad de Santiago de Chile, Chile) and Gabriel Solar (Universidad de Santiago de Chile, Chile)
Copyright: © 2017 |Pages: 23
DOI: 10.4018/978-1-5225-2053-5.ch009
OnDemand PDF Download:
List Price: $37.50


The results of new experiments on the detection and description of images for an EKF-SLAM monocular application are employed in order to obtain a dispersed set of features without related data association problems. By means of different detectors/descriptors, the number of features observed and the ability to observe the same feature in various captures is evaluated. To this end, a monocular vision system independent of the EKF-SLAM system is designed and implemented using the MatLab software. This new system allows for—in addition to image capture—the detection and description of features as well as the association of data between images, thus serving as a priori information to avoid incorrect associations between the obtained features and the map of an EKF-SLAM system. Additionally, it enables the evaluation and comparison of the precision, consistency and convergence of various algorithms.
Chapter Preview

At present, there is a wide range of algorithms for the detection and description of image features, therefore a comparative study of those algorithms becomes necessary to evaluate their behavior in a SLAM system. Although the comparative work of these technologies focuses on artificial vision applications, three important studies in the field of visual localization and SLAM should be noted. Firstly, the work of Gauglitz et al. (2011) presents an extensive analysis of the various combinations between detection and description algorithms for a visual localization application. Secondly, Gil et al. (2010) conduct a comparative study on detectors and descriptors for visual SLAM, which is based on the measurement of the repetition in which features appear in successive images (recall) and on the precision of the features obtained by the compared algorithms. However, despite the comparison being useful to improve the system’s precision, this study does not consider the uncertainty inherent to SLAM. Finally, Klippenstein and Zhang’s (2009) work proposes a new methodology that employs various detectors in order to evaluate the performance of a SLAM system by means of a test of accumulated consistency and uncertainty.

Complete Chapter List

Search this Book: