Online Self-Supervised Thermal Water Segmentation for Aerial Vehicles

TL;DR

This paper introduces an online self-supervised method for thermal water segmentation in aerial imagery, enabling night-time autonomous navigation and environmental monitoring with limited thermal data.

Contribution

The authors develop a novel online self-supervised approach that adapts RGB-trained water segmentation models to thermal imagery using texture and motion cues, and curate the first near-shore thermal dataset.

Findings

Outperforms fully-supervised models trained on limited thermal data

Enables real-time thermal water segmentation onboard embedded platforms

Provides the first dataset for aerial near-shore thermal imagery

Abstract

We present a new method to adapt an RGB-trained water segmentation network to target-domain aerial thermal imagery using online self-supervision by leveraging texture and motion cues as supervisory signals. This new thermal capability enables current autonomous aerial robots operating in near-shore environments to perform tasks such as visual navigation, bathymetry, and flow tracking at night. Our method overcomes the problem of scarce and difficult-to-obtain near-shore thermal data that prevents the application of conventional supervised and unsupervised methods. In this work, we curate the first aerial thermal near-shore dataset, show that our approach outperforms fully-supervised segmentation models trained on limited target-domain thermal data, and demonstrate real-time capabilities onboard an Nvidia Jetson embedded computing platform. Code and datasets used in this work will be available at: https://github.com/connorlee77/uav-thermal-water-segmentation.