physfusion: A Transformer-based Dual-Stream Radar and Vision Fusion Framework for Open Water Surface Object Detection

Researchers have developed PhysFusion, a new AI framework that combines radar and camera data to improve object detection on water surfaces for unmanned vessels. The system achieves up to 94.8% accuracy by using physics-informed processing to handle challenging maritime conditions like wave clutter and poor visibility.

• →PhysFusion framework combines 4D radar and vision data using transformer architecture for maritime object detection.
• →The system achieves 94.8% mAP50 accuracy on FLOW dataset and 90.3% on WaterScenes with efficient 5.6M parameters.
• →Physics-Informed Radar Encoder addresses sparse radar data challenges through scattering priors and reliability prediction.
• →Dual-stream backbone includes point-based local processing and transformer-based global reasoning with specialized attention mechanisms.
• →Temporal Query Aggregation module ensures consistent object tracking across multiple frames for robust detection.