An improved kernel correlation filter for complex scenes target tracking

Video tracking technology employed to achieve efficient and accurate tracking of targets in complex scenes has often been one of the challenges to be tackled. When the target is in a complex scene similar to target interference, it will still create a series of issues, such as template drift althoug...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications 2022-06, Vol.81 (15), p.20917-20944
Main Authors: Huo, Wenxiao, Yan, Yejin, Zhou, Maoxia, Li, Tianping
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Computer Communication Networks
Computer Science
Correlation
Data Structures and Information Theory
Datasets
Kernels
Multimedia Information Systems
Regularization
Robustness
Special Purpose and Application-Based Systems
Tracking
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Video tracking technology employed to achieve efficient and accurate tracking of targets in complex scenes has often been one of the challenges to be tackled. When the target is in a complex scene similar to target interference, it will still create a series of issues, such as template drift although the current target tracking has achieved quality results in terms of accuracy, robustness, and speed. We propose an improved kernel correlation filter algorithm in response to this problem. We introduced a regularization matrix and fused properties of HOG and CN to train an improved kernel correlation filter. Furthermore, an independent scale filter is employed to regulate the scale adaptively. We have introduced a re-detection module to prevent the issue of the kernel correlation filter algorithm relying mainly on the maximum response value.A considerable number of experiments have been conducted on the aforementioned improvements. The algorithm’s average tracking accuracy can attain 85.8%, in the OTB2015 dataset and its running speed can attain 198FPS. The algorithm’s EAO, accuracy, and robustness, on the VOT2016 dataset, can attain 0.303, 0.553, and 0.932, respectively.Experiments demonstrate that our algorithm has satisfactory accuracy and robustness, and satisfies the real-time effect.
ISSN:1380-7501
1573-7721
DOI:10.1007/s11042-022-12669-7