Yuhan Liu

Yuhan Liu

Associate Professor

SiamGM - Satellite Video Tracking

Published:

SiamGM: Siamese Geometry-Aware and Motion-Guided Network for Real-Time Satellite Video Object Tracking
Z Wen, Z Yang, J Li, X Xiang, G Zhou, Y Hu, Y Liu
arXiv preprint, 2026
arXiv

Project Overview

Challenge: Satellite Video Tracking

Satellite video object tracking faces unique challenges:

  • Small target size with limited texture information
  • Rapid motion and scale variations
  • Complex background clutter
  • Real-time processing requirements

Solution: Geometry-Aware Motion-Guided Network

SiamGM addresses these challenges through:

  1. Geometry-aware feature extraction leveraging satellite imaging geometry
  2. Motion-guided target localization using motion patterns
  3. Real-time processing architecture for practical deployment

Key Innovations

1. Geometry-Aware Feature Learning

Incorporates satellite-specific geometric constraints:

  • Camera model awareness
  • Ground plane geometry
  • Scale-space relationships

2. Motion-Guided Localization

Leverages motion cues for robust tracking:

  • Motion pattern prediction
  • Trajectory consistency
  • Dynamic motion model adaptation

3. Real-Time Architecture

Efficient network design:

  • Lightweight backbone
  • Efficient correlation operations
  • Fast feature extraction

Method

Network Architecture

Template Frame + Search Frame
    ↓
[Shared Feature Extraction]
├── CNN backbone
├── Geometry-aware enhancement
└── Multi-scale features
    ↓
[Motion-Guided Matching]
├── Motion prediction
├── Geometry-constrained search
└── Response fusion
    ↓
Target Location + Scale

Results

Performance on satellite video benchmarks:

  • Real-time processing capability
  • Superior tracking accuracy
  • Robust to scale and motion variations

Applications

  • Wide-area surveillance
  • Traffic monitoring from space
  • Maritime vessel tracking
  • Disaster response coordination

Citation

@article{wen2026siamgm,
  title={SiamGM: Siamese Geometry-Aware and Motion-Guided Network for Real-Time Satellite Video Object Tracking},
  author={Wen, Z and Yang, Z and Li, J and Xiang, X and Zhou, G and Hu, Y and Liu, Y},
  journal={arXiv preprint arXiv:2503.07564},
  year={2026}
}