摘要
为了解决目标跟踪中常见的尺度变换、相似目标、背景嘈杂等问题,提出了自适应尺度的上下文感知相关滤波跟踪算法。针对以上问题,在相关滤波跟踪算法的基础上将目标周围的上下文信息作为硬负样本引入分类器中学习,强化分类器的判别能力;通过尺度池在线学习判别式尺度滤波器,在目标位置估计最佳目标尺寸;通过图像帧差均值来评估目标状态并自适应调整模型更新的学习率。实验结果表明提出的算法在快速运动、目标形变等场景下鲁棒性较好。
In order to solve scale transformation,similar target,occlusion,background noisy and other common problems in target tracking,an adaptive scale context-aware correlation filter tracking algorithm is proposed.In view of the above problems,based on the correlation filter tracking algorithm,the context information around the target is introduced into the classifier as a hard negative sample to enhance the discriminative ability of the classifier.The discriminant scale filter is learned online through the scale pool at the target position and estimate the optimal target size.The target state is evaluated by the mean frame difference and the learning rate of the model update is adaptively adjusted.The experimental results show that the proposed algorithm is robust in fast motion,target deformation and any other scenarios.
作者
茅正冲
陈海东
MAO Zhengchong;CHEN Haidong(School of Internet of Things Engineering,Jiangnan University,Wuxi,Jiangsu 214122,China)
出处
《计算机工程与应用》
CSCD
北大核心
2021年第3期168-174,共7页
Computer Engineering and Applications
基金
国家自然科学基金(61701197)。
关键词
尺度变换
上下文感知
相关滤波
帧差均值
scale transformation
context-aware
correlation filter
mean frame difference
