SIA OpenIR  > 光电信息技术研究室
Alternative TitleObject Tracking based on Deep Learning
Thesis Advisor罗海波
Keyword目标跟踪 计算机视觉 深度学习 尺度估计
Call NumberTP391.41/X78/2018
Degree Discipline模式识别与智能系统
Degree Name博士
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract本文以深度神经网络为基础,旨在提出一种即拥有深度学习强大的表达能力,又能满足目标跟踪问题的实时性和鲁棒性需求的跟踪算法。全文贡献如下: 本文总结了目标跟踪算法的研究现状以及其研究意义,尤其在基于深度神经网络的目标跟踪算法上进行了展开了深入探讨和总结;针对于核相关滤波的跟踪算法的不足,设计了一种基于频域分析的尺度估计算法和目标运动状态估计方法。该算法提高了核相关滤波跟踪器在目标发生快速运动和目标发生模糊变化时跟踪算法的精确度和鲁棒性,同时通过有效地估计运动目标的尺度,提升了跟踪算法的性能;为了充分利用深度神经网络的强大学习能力和相关跟踪器良好的实时特性,我们设计了一种带尺度估计和遮挡机制的端到端目标跟踪算法,大幅度提高了目标跟踪算法的精确度和鲁棒性;在第三步的基础上,我们提出了一种在线融合特性方法,将目标的运动信息与物体外观特性融合在一起,有效的提升了算法在对快速物体跟踪的有效性。
Other AbstractBased on the above considerations, this paper is based on a deep neural network and aims to propose a tracking algorithm that possesses a powerful ability of deep learning and can meet the real-time and robust requirements of the target tracking problem. The full text contribution is as follows: This paper summarizes the research status of the target tracking algorithm and its research significance. In particular, it has been deeply discussed and summarized on the target tracking algorithm based on deep neural network; Aiming at the deficiencies of the kernel-related filtering algorithm, a scale estimation algorithm based on frequency domain analysis and a target motion state estimation method are designed. Improves the robustness of the nuclear correlation filtering tracker in the processing of fast moving targets and the occurrence of fuzzy transformation target tracking; In order to make full use of the powerful learning ability of deep neural network and the good real-time characteristics of nuclear correlated tracker, we designed an end-to-end target tracking algorithm with scale estimation and occlusion mechanism, which greatly improved the accuracy of target tracking. Degree and robustness; Based on the third step, we propose an online fusion feature method that fuses the motion information with the appearance characteristics of the object, effectively improving the effectiveness of the algorithm in tracking fast objects.
Contribution Rank1
Document Type学位论文
Recommended Citation
GB/T 7714
许凌云. 基于深度学习的目标跟踪算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2018.
Files in This Item:
File Name/Size DocType Version Access License
基于深度学习的目标跟踪算法研究.pdf(25701KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[许凌云]'s Articles
Baidu academic
Similar articles in Baidu academic
[许凌云]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[许凌云]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.