摘要
针对基于稀疏表示的目标跟踪中编码系数采用L_0或L_1范数正则,易造成NP难优化或预估偏差增大等问题,提出一种基于贝叶斯框架下的组合范数正则化稀疏编码和自适应加权残差的鲁棒跟踪算法.首先提出组合范数正则化稀疏编码,对目标函数编码系数同时进行L_0和L_1正则,根据其贡献程度赋予不同的权值,以增强目标外观模型的鲁棒性;其次在目标函数中引入残差项,赋予其自适应权重来缓解噪声、腐蚀和光照等离群子干扰;最后求解目标函数最小化所涉及的非凸病态问题,在加速近邻梯度算法框架下提出一种广义阈值法来迭代求解最优值.采用大量具有挑战性的序列进行实验的结果表明,与现阶段其他主流算法相比,该算法具有更好的鲁棒性.
In the visual tracking algorithm based on sparse representation,the objective function which used L0 penalty often resulted in NP-hard problem and which used L1 penalty was easy to cause biased estimation.This paper proposes a robust visual tracking method with combined norm regularized sparse coding under the Bayesian inference framework.First,we regularize the coding coefficient with the combined L0 and L1 penalty,and give different parameters values according to their contribution to enhance the robustness of visual tracking.Second,we introduce an adaptive weighted value which considered in the objective function to handle the occlusion,illumination and corrosion problems.Finally,since the minimization of objective function is a non-convex optimization ill-posed problem,we propose a generalized thresholding method under the accelerated proximal gradient method framework to obtain the optimal solution iteratively.Experiments on multiple challenging sequences demonstrate that our tracking method can be much more robust compared with other state-of-the-art methods.
作者
孔军
成静
蒋敏
柳晨华
顾晓峰
Kong Jun;Cheng Jing;Jiang Min;Liu Chenhua;Gu Xiaofeng(Key Laboratory of Advanced Process Control for Light Industry(Ministry of Education),Jiangnan University,Wuxi 214122;College of Electrical Engineering,Xinjiang University,Urumqi 830046)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2018年第4期634-641,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61362030
61201429)
中国博士后科学基金(2015M581720
2016M600360)
科技援疆专项计划(2017E0279)
关键词
组合范数正则编码
自适应权重
目标跟踪
贝叶斯框架
combined norm regularized coding
adaptive weighted value
visual tracking
Bayesian framework
