摘要
针对强遮挡导致的跟踪目标失效问题,提出一种基于字典学习改进的时空上下文算法.先在目标和上下文区域构建前景字典和上下文字典,再利用稀疏解的特性,给提取目标特征更高的权重,并参与模板的更新,构造新的条件概率.实验结果表明,在出现严重遮挡的数据集中,时空上下文算法跟踪成功率为19.5%,改进算法成功率达94.5%,改进算法能在出现强遮挡情况下有效对抗遮挡问题,稳定跟踪.
Aiming at the problem of failure of tracking target caused by strong occlusion,we proposed an improved spatio-temporal context algorithm based on dictionary learning.Firstly,foreground dictionary and context dictionary were constructed in the target area and the context area,and then we used the characteristics of the sparse solution to give higher weight to the feature of the extraction target,the feature with good effect was used to update the template and construct a new conditional probability.The experimental results show that the tracking success rate of spatio-temporal context algorithm is 19.5%in the data set with severe occlusion,and that of the improved algorithm is 94.5%.The improved algorithm can effectively resist the occlusion problem and track stably in the case of strong occlusion.
作者
张尧
才华
曹露
王冰雪
陈广秋
ZHANG Yao;CAI Hua;CAO Lu;WANG Bingxue;CHEN Guangqiu(School of Electronic Information Engineering,Changchun University of Science and Technology,Changchun 130022,China;Changchun China Optics Science and Technology Museum,Changchun 130117,China)
出处
《吉林大学学报(理学版)》
CAS
北大核心
2019年第6期1442-1448,共7页
Journal of Jilin University:Science Edition
基金
国家自然科学基金(批准号:11275046)
吉林省科技发展计划项目(批准号:20170203005GX)
关键词
目标跟踪
稀疏解
时空上下文算法
模板更新
target tracking
sparse solution
spatio-temporal context algorithm
template update