A Normalized Measurement Vector Model for Enhancing Localization Performance of 6-DoF Bearing-only SLAM Sukchang Yun, Yeonjo Kim, Byoungjin Lee, and Sangkyung Sung*
International Journal of Control, Automation, and Systems, vol. 16, no. 2, pp.912-920, 2018
Abstract : "This study proposes a novel bearing measurement model in order to improve the localization performance
of 6-DoF SLAM (six degree-of-freedom simultaneous localization and mapping). The main limitation of the existing
measurement model for 6-DoF bearing-only SLAM using feature points was first analyzed, and a bearing
measurement normalization method was then presented in order to cope with this limitation. The existing measurement
model has a vulnerability in that the bearing measurement has different error levels depending on the feature
point position, and thus the validity of the model is degraded as the feature point moves closer to the origin in the
image. This problem can cause the innovation vector to become abnormally large in extended Kalman filter (EKF)-
based navigation filters, resulting in divergence of the navigation filter. The normalization method proposed in this
study makes the measurement error level constant. The new measurement model was derived using this method,
and a bearing-only SLAM consisting of an inertial measurement unit (IMU) and bearing sensors was constructed
in the EKF framework. The validity of this measurement model was analyzed by checking the innovation vectors
in the navigation filter, and the performance of the system was verified through simulations by comparing with the
navigation solution based on the existing measurement model."
Keyword :
Feature points, model validation, SLAM, vision-based navigation.
Download PDF : Click this link
|