Visual odometry parametres optimization for autonomous underwater vehicles

Visual odometry algorithms are used in a wide range of applications to provide reliable estimates of the movement of the camera frame. Combined with inertial measurement units and other low-cost sensors, they can be used as input of high-level algorithms like Kalman filters or SLAM to estimate the global vehicle pose. However, visual odometers are often complex algorithm pipelines with large parameter sets involved in every process stage. In this paper we focus on optimizing the parameters of a visual odometer that has proved to work in underwater environments. We present the results of a exhaustive optimization process to reduce the errors in rotation and translation comparing vision-based pose estimates with ground truth in real underwater environments.