Bridging the Performance Gap in USV Operations: Confronting Water Surface Reflection Challenges

Unmanned Surface Vessels (USVs) or drone boats are vital for maritime tasks like
environmental monitoring and transport. Several mechanisms have been proposed to
ensure its operational safety. However, the challenges posed by the reflection
phenomenon on the water surface, aka Water Surface Reflection (WSR), have been
largely overlooked. First, the WSR can create dangerous visual ambiguities. Such
ambiguities generate false targets that could mislead USVs, potentially resulting in
catastrophic collisions. Second, the WSR occludes critical details of the maritime
landscape, blinding USVs to objects of interest. The goals of this project is to mitigate
the visual ambiguities and occlusion caused by WSR, and ultimately enhance the
reliability of USV operations.
Two key challenges hinder current approaches from achieving the desired goals. First,
USVs rely on integrating 2D and 3D data for effective operation. However, most
reflection-related approaches focus solely on 2D data, such as images or videos, withoutaccounting for spatiotemporal correlations in a 3D environment. This limitation leads to
suboptimal outcomes, significantly affecting USV performance. Second, current methods
are primarily designed for stable glass reflections. In contrast, water surface reflections
in maritime environments change rapidly across different locations and times, making
these methods ineffective in handling such swift variations.
This project will address the challenges from three aspects. First, we will introduce a
novel approach to model WSR via a base prior with 3D particle structure. The base prior
segments the scene into a collection of small particles, each capable of modeling
spatiotemporal variations at its position by fine-tuning associated trainable parameters.
This base prior will then be continuously updated, ensuring accurate and timely
modeling of WSR. Second, building on such a base prior, we will explore how to conduct
in-prior reflection handling to control the ambiguities and occlusion effects in 3D. At
last, we will develop a framework to seamlessly incorporate base prior into existing USV
systems. Since our base prior is built on 3D particle structures, it naturally delivers the
necessary information for USV operations.
Our extensive experience and strong research teams in maritime tasks and
computational imaging give us confidence in addressing WSR challenges via this project.
This research will significantly enhance USV safety. The scientific advances from this
three-year project are expected to have broad and lasting impacts beyond the initial
timeframe.