Visual-Inertial Odometry for Metric-Scale Mapping of Underwater Caves

Mapping underwater caves is a challenging endeavor. Complex morphology and fragile geological formations limit the use of tethered ROV or AUV platforms for data collection. Trained cave divers remain the most effective means to achieve this task. The convenience of using small-size monocular cameras implies Structure-from-Motion (SfM) reconstructs 3D caves lacking a metric scale factor. Combining SfM with Visual Inertial Odometry (VIO) offers the potential for metrically scaling cave mapping. This work proposes a full pipeline for scaled 3D reconstruction of under-water caves, including dataset acquisition and sensor calibration methodologies, and the combined use of VIO and SfM for metric trajectory estimation. The proposed approach is validated with real visual-inertial datasets created by cave divers with a GoPro Hero 9 camera, equipped with an IMU, processed through an Unscented Kalman Filter (UKF) which allows estimating pose in metric units.