光学 精密工程, 2018, 26 (2): 492, 网络出版: 2018-03-21   

利用最佳伙伴相似性的改进空间正则化判别相关滤波目标跟踪

Improved SRDCF object tracking via the Best-Buddies Similarity
作者单位
河北工业大学 控制科学与工程学院,天津 300130
摘要
针对空间正则化判别相关滤波跟踪算法(SRDCF)在目标发生遮挡、尺度变化和形变情况下的跟踪失败问题,提出利用最佳伙伴相似性的改进SRDCF目标跟踪算法。首先,以SRDCF算法为基础,利用双层搜索策略解决目标跟踪中的目标定位问题和尺度估计问题;然后,利用一种新颖的鲁棒模板匹配技术,通过融合空间权重、相关滤波得分和最佳伙伴相似性得分来估计候选目标位置,解决遮挡情况下的目标重定位问题;最后,采用自适应模板更新策略解决遮挡情况下模板漂移问题。本文采用OTB-2013数据集评估本文算法的性能,同时与34种流行算法进行比较,结果表明本文算法的精确度得分和成功率得分分别为0.853和0.648,相比传统的SRDCF算法分别提高1.79%和3.51%。本文算法能很好地解决目标遮挡、尺度变化和形变情况下的目标跟踪问题,具有一定研究价值。
Abstract
Aiming at the failure of tracking via spatially regularized discriminant correlation filter (SRDCF) algorithm caused by occlusion, scale change and deformation, an improved SRDCF algorithm based on Best-Buddies Similarity was proposed. Firstly, the proposed algorithm based on SRDCF, locating target and estimating scale in the process of object tracking were complemented by using bi-level search strategy. Secondly, a novel robust template matching technique was used to estimate the candidate object position by integrating the spatial weights, the correlation filter score and the Best-Buddies Similarity score, thus the problem of target relocation in the occlusion was resolved. Finally, the adaptive template updating strategy was employed to mitigate the template drift problem in the case of occlusion. The performance of the proposed algorithm was evaluated on OTB-2013 datasets and was compared with 34 popular algorithms. The results show that the accuracy and the success rate of the proposed algorithm are 0.853 and 0.648, which are 1.79% and 3.51% higher than the traditional SRDCF algorithm, respectively. The proposed algorithm can deal with the matter of occlusion, scale change and deformation effectively, and has some value of research.

杨德东, 毛宁, 杨福才, 李雪晴. 利用最佳伙伴相似性的改进空间正则化判别相关滤波目标跟踪[J]. 光学 精密工程, 2018, 26(2): 492. YANG De-dong, MAO Ning, YANG Fu-cai, LI Xue-qing. Improved SRDCF object tracking via the Best-Buddies Similarity[J]. Optics and Precision Engineering, 2018, 26(2): 492.

本文已被 8 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!