Inter-Camera Tracking Based on Fully Unsupervised Online Learning

• Paper
• Poster

Summary

This method associates people across non-overlapping cameras through fully unsupervised online learning. A two-stage feature extractor combines appearance and contextual information, including automatically detected companion pairs, while a camera-link model learns relationships between views. Experiments on NLPR_MCT evaluate cross-camera tracking performance and the scalability of the approach without manually labeled camera-link data.

Citation

@inproceedings{Lee17UnsupervisedICT,
  author = {Young-Gun Lee and Zheng Tang and Jenq-Neng Hwang and Zhijun Fang},
  title = {Inter-camera tracking based on fully unsupervised online learning},
  booktitle = {Proc. ICIP},
  pages = {2607--2611},
  address = {Beijing, China},
  year = {2017}
}