Multi-frame visual odometry in image-aided inertial navigation system

Abstract

This paper presents a novel stereo image-based image aided inertial navigation algorithm for reducing position and orientation drifts during GNSS outages or in a poor GNSS environment. Usually, the image aided navigation based on the visual odometry uses the tracked features only from a pair of the consecutive image frames. The proposed method integrates the features tracked from all overlapping image frames towards accuracy improvement and drift reduction. The measurement equation system in this multi-frame visual odometry algorithm (MFVO) is derived from Simultaneous Localization and Mapping (SLAM) measurement equation system where the landmark position parameters in SLAM are algebraically eliminated by time-differencing the measurement at two consecutive epochs. However the resulted time-differenced measurements are time-correlated. Through a sequential de-correlation the Kalman filter measurement update can be performed sequentially and optimally. Monte Carlo simulations show that the MFVO and SLAM pose estimates are similar. Compared with SLAM, the proposed method uses less computation resources especially when the number of features in view is large. The results from a real dataset are also presented.