Multiple-Kernel Adaptive Segmentation and Tracking (MAST) for Robust Object Tracking

• Paper
• Slides

Summary

MAST improves object segmentation and tracking when foreground objects resemble the background or move through shadows. An initial multiple-kernel tracker provides feedback that adapts background-subtraction and shadow-removal thresholds around each target. A second tracking pass uses the refined foreground masks. Evaluations on public benchmarks and recorded video examine the effect of this feedback loop on segmentation quality and tracking robustness.

Citation

@inproceedings{Tang16MAST,
  author = {Zheng Tang and Jenq-Neng Hwang and Yen-Shuo Lin and Jen-Hui Chuang},
  title = {Multiple-kernel adaptive segmentation and tracking ({MAST}) for robust object tracking},
  booktitle = {Proc. ICASSP},
  pages = {1115--1119},
  address = {Shanghai, China},
  year = {2016}
}