摘要
为解决目标跟踪中目标形变、遮挡等因素导致目标外观大幅度变化的问题,提出了记忆驱动的空时相关滤波跟踪算法。首先使用交叉熵公式度量前、后两帧模型的差异,以确定样本的置信度;然后通过置信度储存跟踪目标的外观记忆,并使用外观记忆对模型做时间上的约束,以增加跟踪模型的抗干扰性。基于公开数据集OTB2015进行算法性能测试,结果显示,所提出的目标跟踪算法的跟踪精度和跟踪成功率皆有所提升,尤其是对目标遮挡、形变类视频的跟踪效果提升显著。
In order to address the problem of object tracking where the target undergoes significant appearance variation due to deformation or occluded,a memory-driven spatial-temporal correlation filter for object tracking was proposed.Cross entropy formula was utilized to measure the difference between previsous model and current model for determining the confidence coefficient of sample in each frame.This algorithm stored target appearance memory by cofidence coefficient.A temporal regularization term composed of target appearance memory was applied to the original model for anti-interference performance.Public dataset OTB2015 was used to test the algorithm.Experimental results show that the proposed algorithm has leading performance on distance precision and success rate,especially in the attributes of occlusion and deformation.
作者
孙浩
韩立新
徐国夏
SUN Hao;HAN Lixin;XU Guoxia(College of Computer and Informaion,Hohai University,Nanjing 211100,China)
出处
《中国科技论文》
CAS
北大核心
2019年第7期711-717,共7页
China Sciencepaper
关键词
图像处理
目标跟踪
相关滤波
空时约束
外观记忆
image processing
object tracking
correlation filter
spatial-temporal regularization
appearance memory
