For an autonomous vessel, navigation cannot simply mean knowing where it is and where it is heading. It must understand its surroundings, identify and track other vessels and objects, and distinguish between them.
Under some operating conditions, radar and automatic identification system (AIS) may provide an incomplete or ambiguous picture, while operational requirements may call for active sensors to be limited or switched off altogether. Electro-optical sensing therefore provides an important layer of resilience within a multi-modal sensor stack, enabling USVs to passively detect, track, and classify contacts using only their optical systems.
Building the Operational Picture
This is where AI-enabled processing can play an important role. At Chess Dynamics, maritime electro-optical systems such as SeaEagle work alongside Vision4ce’s video tracking, image processing and AI capabilities to help platforms build a more reliable picture of their surroundings.
At Chess Dynamics, we have invested significant research and development into track-while-scan technology for electro-optical directors. The capability brings together advances in object detection, multi-target tracking and fine-grained classification, allowing a sensor to periodically scan a wide area while detecting and maintaining tracks on multiple contacts. Unlike conventional video tracking, targets may disappear from view for significant periods as the sensor continues its scan. The challenge is therefore detecting a target, but also recognizing and classifying it when the sensor returns, correctly associating it with its previous track and building a persistent picture of activity over time.
Granular classification is particularly important for autonomous operations, providing vital context for decision-making. Our COLREGs classification model can distinguish between vessel types and sizes, including small, medium and large vessels, sailing vessels and navigation aids, as well as potential threats, supporting navigation and broader threat detection.
Processing at the Edge
But algorithms are only part of the solution. USVs have finite space, power and processing resources making low size, weight and power (SWaP) and low latency essential.
In contested, congested or remote environments, bandwidth may be limited and communications links intermittent. Edge processing allows vessels to interpret sensor data and maintain situational awareness independently, while sharing information with the wider force when communications are available.
Developments such as CHARM50 build on Chess Dynamics’ experience in real-time edge video processing, delivering increasingly sophisticated algorithms within the space, power and compute constraints of modern platforms. This requires careful optimization to balance computational efficiency with performance.
Building Confidence in Autonomy
As USVs move into more demanding operations alongside crewed vessels, operators need confidence that uncrewed platforms can provide an accurate assessment of their environment and that this information can be acted upon across the wider fleet.
By combining AI-enabled identification, tracking and classification with low-SWaP edge video processing, USVs can provide crucial intelligence, surveillance and reconnaissance (ISR) data to support their own survivability and that of the wider fleet.
