Wide Baseline Matching for Autonomous Approaches of MAVs

Karsten Mueller, Ruben Kleis, Gert F. Trommer

Abstract:	A method for detecting targets for autonomous approaches to buildings of micro aerial vehicles is presented. The selected target is a window marked in a reference image which was recorded from a different perspective. We propose a wide baseline matching algorithm based on ORB feature points that offers high accuracy and low processing time. By applying projective transformations to the reference image, invariance to out-of-plane perspective transformations is achieved. Robustness of the homography estimation is improved by filtering keypoint orientation and applying a descriptor ratio test. Moreover, a rating for the estimated homography matrices based on cross correlation values and a color descriptor is introduced. This rating allows for an accurate and robust selection of the correct homography estimate. Since not only pairs of images but image sequences are available for an autonomous approach to buildings, tracking is included in the algorithm. In order to avoid divergence, a re-initialization procedure is presented. The algorithm achieves a high detection rate and good invariance to large perspective changes while meeting real-time computation requirements.
Published in:	2018 IEEE/ION Position, Location and Navigation Symposium (PLANS) April 23 - 26, 2018 Hyatt Regency Hotel Monterey, CA
Pages:	801 - 808
Cite this article:	Mueller, Karsten, Kleis, Ruben, Trommer, Gert F., "Wide Baseline Matching for Autonomous Approaches of MAVs," 2018 IEEE/ION Position, Location and Navigation Symposium (PLANS), Monterey, CA, April 2018, pp. 801-808.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In