Related Experiment Video
Updated: Nov 18, 2025
13:02
Three-dimensional Particle Tracking Velocimetry for Turbulence Applications: Case of a Jet Flow
Published on: February 27, 2016
12.6K
HRSiam: High-Resolution Siamese Network, Towards Space-Borne Satellite Video Tracking
Summary
This study introduces a novel high-resolution Siamese network (HRSiam) for tracking small, moving objects in satellite videos. HRSiam enhances precision and real-time performance, overcoming challenges like low resolution and occlusion.
Area of Science:
- Computer Vision
- Remote Sensing
- Artificial Intelligence
Background:
- Tracking small objects in satellite videos is challenging due to limited pixels, appearance variations, and occlusion.
- Existing methods often use low-resolution features and neglect inter-frame motion information, hindering accurate localization.
Purpose of the Study:
- To develop a lightweight, high-resolution network for precise and real-time small object tracking in satellite imagery.
- To improve tracking robustness against occlusion and illumination variations using a pixel-level refining model.
Main Methods:
- Designed a lightweight parallel network with high spatial resolution for small object localization.
- Integrated a pixel-level refining model using online moving object detection and adaptive fusion.
- Applied the architecture to Siamese Trackers for enhanced performance.
Main Results:
- The proposed HIGH-RESOLUTION SIAMESE NETWORK (HRSiam) achieves state-of-the-art tracking performance on real satellite video datasets.
- The network demonstrates real-time processing capabilities, running at over 30 frames per second.
- The pixel-level refining model significantly enhances tracking robustness.
Conclusions:
- HRSiam effectively addresses the challenges of small object tracking in satellite videos.
- The proposed method offers a robust and efficient solution for real-time satellite video analysis.
- This work advances the capabilities of object tracking in remote sensing applications.
