Improving visual odometry for challenging underwater environments and AUV Navigation

Abstract

[eng] Due to the need of having one more sensor to calculate the location of an Autonomous Underwater Vehicle (AUV), in order to increase the accuracy in estimating its position and velocity, it has been decided to improve the performance of the well-known LIBVISO2 visual odometer in underwater environments colonized by algae or seagrass meadows. Furthermore, increasing its robustness to rotations and scale changes has also been one of the main objectives of this work. To achieve these targets, its feature detection and matching blocks have been modified. First, the detection block has been replaced by one that is more robust to the complex textures of these environments and also invariant to rotations and scale changes. Secondly, the efficiency of the matching block has been improved to reduce the number of outliers produced by these textures.