Keyframe Selection for Acoustic Odometry Estimation Using Forward Looking Sonar

This study proposes a method to improve the accuracy of acoustic odometry using keyframe selection. The acoustic odometry was estimated based on Fourier-based image registration using images acquired from forward-looking sonar (FLS). The odometry estimation error increases when image registration fails due to spurious peaks and high similarity image pairs. The image pair selection must be adjusted prior to image registration in order to reduce the effects of spurious peaks and high similarity image pairs. We proposed a method for the adaptive frame interval by analyzing the factors affecting phase correlation for the accuracy of odometry. The odometry errors were reduced by using keyframe selection using adaptive frame intervals. The proposed method of verification using real-sea data is presented by comparing the odometry estimation results of USBL and a fixed frame interval using UUV cyclops.