Long Time Target Tracking Algorithm Based on Multi Feature Fusion and Correlation Filtering
Long Time Target Tracking Algorithm Based on Multi Feature Fusion and Correlation Filtering
复制标题
DOI：
10.1145/3488933.3488996
复制
发表时间：
2021-09
期刊：
影响因子：
--
通讯作者：
Junsuo Qu;Yuan Zhang;Kai Zhou;Abolfazl Razi
中科院分区：
文献类型：
--
作者：
Junsuo Qu;Yuan Zhang;Kai Zhou;Abolfazl Razi
本文研究了复杂场景中由于光照变化、目标变形、尺度变化、运动模糊等因素导致跟踪失败的情况下的长期目标跟踪问题。更具体地说，我们提出了一种基于尺度自适应核相关滤波和再检测融合的目标跟踪算法，称为再检测多特征融合算法(RDMF)。目标跟踪算法首先训练基于HOG、CN和LBP特征的三个核相关滤波器，然后根据APCE准则得到不同特征对应的响应图的融合权值，并利用加权平均完成对被跟踪目标的位置估计。针对跟踪过程中目标被遮挡和消失的问题，训练一个随机蕨类分类器，在目标被遮挡时进行重新检测。通过对OTB-50目标跟踪数据集的比较，RDMF算法比SAMF算法提高了10.1%的测距精度，优于KCF、DSST、CN等算法。
This paper considers the problem of long-term target tracking in complex scenes when tracking failures are unavoidable due to illumination change, target deformation, scale change, motion blur, and other factors. More specifically, we propose a target tracking algorithm, called Re-detection Multi-feature Fusion (RDMF), based on the fusion of Scale-adaptive kernel correlation filtering and re-detection. The target tracking algorithm trains three kernel correlation filters based on HOG, CN and LBP features, and then obtains the fusion weight of response graphs corresponding to different features based on APCE criterion, and uses weighted Average to complete the position estimation of the tracked target. In order to deal with the problem that the target is occluded and disappears in the tracking process, a random fern classifier is trained to perform re-detection when the target is occluded. After comparing the OTB-50 target tracking data set, the RDMF algorithm improves the range accuracy by 10.1% compared with SAMF algorithm, and is better than KCF, DSST, CN and other algorithms.