Joint Multi-View People Tracking and Pose Estimation for 3D Scene Reconstruction

• Paper
• Slides

Summary

This framework jointly tracks people and estimates their 3D poses across overlapping camera views. It associates single-camera tracklets using appearance, geometry, and pose or action cues, then selects informative viewpoints to reconstruct each person’s 3D skeleton. Estimated joint positions feed back into tracking to improve association. Experiments evaluate the approach on multi-view tracking and 3D pose-estimation benchmarks.

Citation

@inproceedings{Tang18JointTrackHPE,
  author = {Zheng Tang and Renshu Gu and Jenq-Neng Hwang},
  title = {Joint multi-view people tracking and pose estimation for {3D} scene reconstruction},
  booktitle = {Proc. ICME},
  address = {San Diego, CA, USA},
  year = {2018}
}