Quaternion-Based Robust PCA for Efficient Moving Target Detection and Background Recovery in Color Videos

Hi! This is the repository of uQRPCA+. Thank you very much for your interest in my work.
If you have any suggestions, feedback, or are interested in potential collaboration, feel free to reach out 📧. I'm always happy to connect and learn from others in the community.

One of our main goals is to serve supervised deep learning methods: the data generated by our framework can be used as a weakly supervised paradigm in neural networks. While you may choose to assign it a lower loss weight during training, it might significantly improve generalization performance in real-world scenarios.

Co-authors: Liyang Wang, Shiqian Wu, Shun Fang, Qile Zhu, Jiaxin Wu, Sos Agaian

---

🔧 uQRPCA+: Universal Quaternion-Based Robust PCA with Color Rank-1 Batch

A universal and efficient method for background modeling and moving target detection in color videos, based on quaternion robust PCA.

---

📰 News

• [2025/] 🔧 uQRPCA+ code coming soon!
• [2025/07/29] 📄 The paper is now available on arXiv, and the dataset has been released.
• [2025/07/09] 🧪 Evaluation code released.

---

🚀 Demo & Main Scripts

coming soon!

https://www.ece.nus.edu.sg/stfpage/eleclf/ https://www.ece.nus.edu.sg/stfpage/eleclf/Block%20RPCA%20matlab.rar

---

📊 Evaluation

🎞 Background Recovery

• Platform: MATLAB
• Scripts: main.m

🎯 Moving Target Detection

• Platform: PYTHON
• Scripts: main.py, scale_change.py

---

📄 Citing

If you find this work useful in your research, please consider citing:

@misc{wang2025uqrpca,
  title        = {Quaternion-Based Robust PCA for Efficient Moving Target Detection and Background Recovery in Color Videos},
  author       = {Wang, Liyang and Wu, Shiqian and Fang, Shun and Zhu, Qile and Wu, Jiaxin and Agaian, Sos},
  year         = {2025},
  eprint       = {2507.19730},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV}
}