Real-time multi-candidates fusion based head tracking on Kinect depth sequence

Zhiting Yang; Yang Yang; Yunxia Liu

ICASSP 2016accepted0 citations

Real-time multi-candidates fusion based head tracking on Kinect depth sequence

Abstract

Considering depth images are robust to illumination variations with complex backgrounds, the paper developed a real-time head tracking system with one Kinect camera. Distance transform is applied to pre-processed depth frames to further reduce the effect of appearance deformation. A multi-candidates fusion strategy is proposed for template updating that assures head representation robustness. Two-stage template matching is adopted for computational efficiency in the searching procedure. In addition, an early termination criterion for template updating is presented to reliably improve the tracking accuracy. Abundant experimental results on our depth database demonstrate that the proposed method performs favorably against state-of-the-art methods in terms of robustness, accuracy, and efficiency.

BibTeX

@inproceedings{icassp2016_realtimemultican,
  title = {Real-time multi-candidates fusion based head tracking on Kinect depth sequence},
  author = {Zhiting Yang and Yang Yang and Yunxia Liu},
  booktitle = {ICASSP 2016},
  year = {2016}
}