基于重要性加权的结构稀疏跟踪方法  

Structural Sparse Tracking Method Based on Importance Weighting

作　　者：梁贵书[1] 牛为华[2] 李宝树[1] 李强 赵鹏[3] LIANG Guishu;NIU Weihua;LI Baoshu;LI Qiang;ZHAO Peng(Department of Electrical Engineering, North China Electric Power University, Baoding Hebei 071003, China;Department of Computer, North China Electric Power University, Baoding Hebei 071003, China;State Grid Hebei Electric Power Company, Shifiazhuang 050021, China)

机构地区：[1]华北电力大学电力工程系,河北保定071003 [2]华北电力大学计算机系,河北保定071003 [3]河北省电力公司,石家庄050021

出　　处：《传感技术学报》2018年第1期61-67,共7页Chinese Journal of Sensors and Actuators

基　　金：中央高校基本科研业务费专项项目(2017MS156)

摘　　要：针对视觉跟踪中描述目标能力的有限性和局部稀疏表示模型的有效性,提出了一种基于重要性加权的结构稀疏跟踪方法。该方法采用结构稀疏表示对目标表观建模,根据在表达目标表观时所起的作用,对每个局部图像进行加权处理;在粒子滤波框架下,应用最大后验概率对目标的状态进行估计;通过带有遮挡检测机制的模板更新策略对目标模板进行在线的更新以避免跟踪漂移。实验表明,该方法有效地减弱了目标表观变化对模型的影响,对于视频序列中的遮挡、光照变化和目标姿态改变等有稳健的跟踪效果。According to the limitations of describe target capabilities and the effectiveness of local sparse representation model in visual tracking,a structural sparse tracking method based on importance weighting is proposed aimed to the deficiencies. In the method,we adopt sparse representation to model object and according to important degree of expressing object,each local image is weighted to improve robustness of target model. In the framework of particle filtering,based on maximum a posteriori probability to estimate target. In addition,a template updating strategy of occlusion detection mechanism is used to real-time update template to avoid tracking drift. Experimental results show that the proposed method can effectively reduce the influence of the object apparent change on the model and our method is competitive to the state-of-the-art trackers on challenging video sequences with illumination changes,background clutter,occlusion,fast motion and deformation.

关 键 词：目标跟踪 稀疏表示 重要性加权 目标表观 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]