[1]梁文辉,宋 涛,叶永达.间隔贝叶斯估计和空时目标运动边界视觉跟踪[J].计算机技术与发展,2021,31(09):75-80.[doi:10. 3969 / j. issn. 1673-629X. 2021. 09. 013]
　LIANG Wen-hui,SONG Tao,YE Yong-da.Visual Tracking Based on Interval Bayesian Estimation and Spatial-temporal Object Motion Boundaries[J].,2021,31(09):75-80.[doi:10. 3969 / j. issn. 1673-629X. 2021. 09. 013]
点击复制

间隔贝叶斯估计和空时目标运动边界视觉跟踪()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
31
期数:
2021年09期
页码:
75-80
栏目:
图形与图像
出版日期:
2021-09-10

文章信息/Info

Title:
Visual Tracking Based on Interval Bayesian Estimation and Spatial-temporal Object Motion Boundaries
文章编号:
1673-629X(2021)09-0075-06
作者:
梁文辉1 ; 宋 涛2 ; 叶永达3
1. 解放军 31004 部队,北京 100094;
2. 解放军信息工程大学,河南 郑州 450000;
3. 解放军陆军工程大学,江苏 南京 210001
Author(s):
LIANG Wen-hui1 ; SONG Tao2 ; YE Yong-da3
1. 31004 Troops of PLA,Beijing 100094,China;
2. Information Engineering University of PLA,Zhengzhou 450000,China;
3. Army Engineering University of PLA,Nanjing 210001,China
关键词:
视觉跟踪; 间隔贝叶斯估计; 目标运动边界; 双层目标外观模型; 模型更新
Keywords:
visual tracking; interval Bayesian estimation; object motion boundaries; two-level object appearance model;  model update
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2021. 09. 013
摘要:
针对计算机视觉领域的运动目标跟踪问题,在间隔贝叶斯估计框架的基础上结合空时目标运动边界提出一种在线视觉跟踪方法。 首先在初始帧建立感兴趣目标的整体和局部双层外观模型;然后利用目标驱动和数据驱动的双层视觉注意力模型提取目标在下一帧内的运动边界,将目标运动边界视为状态预测间隔的上限;在预测间隔内采用多尺度匹配原则寻找最优的目标框选窗口,并利用局部模型判断目标的遮挡状态;根据目标局部子块的匹配权值提出基于交叉限制的在线更新机制,能够长时间保持目标整体模型的同时利用子块实时捕捉目标外观的局部变化。 通过与其他几种具有代表性的跟踪算法的比对实验,验证了该算法在相似背景干扰和尺度变化的应用场景中存在明显优势,能够有效拟合目标边界,跟踪运动目标。
Abstract:
Aiming at the issue of moving object tracking in the field of computer vision,an online visual tracking method is proposed based on the framework of interval Bayesian estimation,combined with the spatial-temporal object motion boundaries. First,a two-level object appearance model is established including the global and local appearance models in the first frame. Then the object motion boundaries,which are considered as the upper bounds of state prediction interval,are extracted in the next frame, using a two-level visual attention model driven by target and data. In the prediction interval, an optical target window is found, employing the multi - scale matching principle and the occlusion state of an object is detected by local model. Finally,a cross-constrained online update mechanism is put forward according to the matching weight of local object sub-blocks. It can keep the overall model of the target for a long time while using sub-blocks to capture the local changes of the target’s appearance in real time. Compared with other representative tracking approaches, experimental results have verified that the proposed algorithm has obvious advantages in application scenarios with similar background interference and scale changes. It can effectively fit the target boundary and track moving targets.

相似文献/References:

[1]王业祥,朱文球,孙文静.局部加权距离度量的双向稀疏表示目标跟踪[J].计算机技术与发展,2018,28(04):60.[doi:10.3969/ j. issn.1673-629X.2018.04.013]
　WANG Ye-xiang,ZHU Wen-qiu,SUN Wen-jing.Target Tracking of Bidirectional Sparse Representation of Local Weighted Distance Metric[J].,2018,28(09):60.[doi:10.3969/ j. issn.1673-629X.2018.04.013]
[2]田应仲,刘伊芳,李 龙.联合判别式核相关滤波视觉跟随算法[J].计算机技术与发展,2021,31(03):78.[doi:10. 3969 / j. issn. 1673-629X. 2021. 03. 013]
　TIAN Ying-zhong,LIU Yi-fang,LI Long.Visual Following Algorithm of Joint Discriminant Kernel Correlation Filter[J].,2021,31(09):78.[doi:10. 3969 / j. issn. 1673-629X. 2021. 03. 013]

更新日期/Last Update: 2021-09-10