Knowledge Distillation for Feature Extraction in Underwater VSLAM

TL;DR

This paper introduces a cross-modal knowledge distillation approach to train underwater feature detection and matching networks, improving underwater VSLAM performance using synthetic data and a new dataset.

Contribution

It proposes a novel knowledge distillation framework for underwater feature extraction and integrates it into VSLAM with a new underwater dataset.

Findings

Effective feature detection in underwater environments demonstrated.

Improved VSLAM accuracy with the proposed method.

Successful use of synthetic data for training underwater features.

Abstract

In recent years, learning-based feature detection and matching have outperformed manually-designed methods in in-air cases. However, it is challenging to learn the features in the underwater scenario due to the absence of annotated underwater datasets. This paper proposes a cross-modal knowledge distillation framework for training an underwater feature detection and matching network (UFEN). In particular, we use in-air RGBD data to generate synthetic underwater images based on a physical underwater imaging formation model and employ these as the medium to distil knowledge from a teacher model SuperPoint pretrained on in-air images. We embed UFEN into the ORB-SLAM3 framework to replace the ORB feature by introducing an additional binarization layer. To test the effectiveness of our method, we built a new underwater dataset with groundtruth measurements named EASI (https://github.com/Jinghe-mel/UFEN-SLAM), recorded in an indoor water tank for different turbidity levels. The experimental results on the existing dataset and our new dataset demonstrate the effectiveness of our method.