Neuromorphic vision breakthrough enables efficient underwater optical flow

Underwater environments are notoriously challenging for conventional cameras due to low light, turbidity, and motion blur. Event cameras, which trigger pixels only on brightness changes, offer a promising alternative with high temporal resolution and low power consumption. However, their application underwater has been largely unexplored. In a new paper on arXiv, researchers Pei Zhang, Yunkai Liang, and Kaiqiang Wang pioneer the use of neuromorphic vision for underwater motion field estimation. They built a self-supervised learning framework based on spiking neural networks (SNNs) that estimates dense optical flow directly from asynchronous event streams. The self-supervised approach elegantly sidesteps the chronic scarcity of annotated underwater datasets, a major bottleneck for traditional deep learning methods.

Extensive evaluations show the method achieves competitive visual and quantitative results compared to leading optical flow techniques, while maintaining superior computational efficiency. Because SNNs process events in a biologically inspired, event-driven manner, they consume far less power and computation than conventional artificial neural networks. This makes the framework ideal for deployment on lightweight, resource-constrained edge platforms such as underwater drones, autonomous underwater vehicles (AUVs), and robotic sensors. The work bridges neuromorphic sensing and aquatic intelligence, promising real-time, low-cost perception for applications like underwater inspection, environmental monitoring, and search-and-rescue operations.