摘要
为解决在复杂背景条件下的跟踪不稳定问题,提高目标跟踪的鲁棒性和准确性,研究一种在传统核相关滤波算法的基础上对多特征进行线性融合和多峰值检测更新机制结合的核相关滤波目标跟踪算法,使用多个专家进行评估,充分结合各特征的优势,训练出最优的相关滤波器。通过OTB-2013公开数据集全部视频序列对算法进行验证,该算法准确度能达到81.7%,成功率达到69.2%,验证了该算法能够在旋转、运动模糊、快速运动、形变、光照变化和超出视野等场景下取得较好的结果,是一种稳定的目标跟踪算法。
To solve the tracking instability problem under complex background conditions and improve the robustness and accuracy of target tracking,based on the traditional kernel correlation filter algorithm,a kernel correlation filter target tracking algorithm combining multi-feature linear fusion and multi-peak detection update mechanism was proposed.The method named multi-cue kernel correlation filter tracking algorithm based on multi-feature fusion(MCMF)fully combined the advantages of various features to train the optimal correlation filter.The algorithm was verified using all video sequences of the OTB-2013 public dataset.The accuracy of the algorithm is 81.7%,and the success rate is 69.2%.The results show that the proposed algorithm can achieve better results in scenes such as rotation,motion blur,fast motion,deformation,lighting change and out-of-view.It is a stable target tracking algorithm.
作者
龚真
王晓凯
陈浩
GONG Zhen;WANG Xiao-kai;CHEN Hao(College of Physics and Electronic Engineering,Shanxi University,Taiyuan 030006,China;School of Astronautics,Beihang University,Beijing 100083,China)
出处
《计算机工程与设计》
北大核心
2021年第3期670-677,共8页
Computer Engineering and Design
基金
山西省重点研发计划(高新技术领域)基金项目(201803D121102)。
关键词
目标跟踪
核相关滤波
特征融合
多峰值检测
多专家评估
target tracking
kernel correlation filtering
feature fusion
multi-peak detection
multi-cues
