Summary
the 2014 International Symposium on Nonlinear Theory and its Applications
2014
Session Number:D2L-B
Session:
Number:D2L-B2
Modified Mean Shift Tracking with Variational Normalization Coefficient
Shaozhuo Zhai, Yuehu Liu, Xinzhao Li, Zhichao Cui,
pp.755-758
Publication Date:2014/9/14
Online ISSN:2188-5079
Summary:
Mean shift algorithm has a promising performance on object tracking due to its simplicity and robustness, while the compromise between standard kernel density estimation and visual target representation degrades the tracking performance. By using anisotropic kernels defined on object shape and introducing kernel orientation in KDE to describe object rotation, we propose a target representation with better precision. Then based on this representation, a modified mean shift iteration procedure is derived by minimizing the distance function with variational normalization coefficient. Experiments demonstrate that the presented algorithm tracks target with more accuracy without increasing the computational complexity compared with classical mean shift tracking algorithm.