IOE OpenIR  > 光电技术研究所博硕士论文
复杂场景下遮挡干扰目标鲁棒跟踪技术研究
崔盼果
Subtype硕士
Thesis Advisor周进 ; 雷涛
2018-06
Degree Grantor中国科学院研究生院
Place of Conferral北京
KeywordKcf Stc 遮挡检测 跟踪难度
Abstract

机器视觉技术日新月异,其重要组成部分——目标跟踪技术已越来越受到重视,且在军事和民用范畴都有广泛普及。然而由于跟踪环境的复杂多变性,如光照变化、目标变形、遮挡、相似目标干扰等影响,在跟踪过程中,跟踪器很容易丢失或者跟错目标,而遮挡干扰是引起跟踪失败的主要原因。

本文的研究内容是对复杂场景中运动目标跟踪算法进行研究和改进,在基于时空上下文(Spatial-Temporal Context,STC)算法和核相关滤波(Kernel Correlation Filters,KCF)算法的基础上,解决遮挡干扰问题,同时设计对视频目标跟踪难度评价指标进行改进,从目标运动过程中的背景因素及目标本身因素进行分析,定量描述视频序列中目标的跟踪难度。针对STC算法和KCF算法中的问题,提出了基于局部二值模式(Local  Binary Pattern, LBP)的改进时空上下文算法(LKSTC)和基于核相关滤波的层级遮挡检测算法,算法跟踪性能明显提升。本文主要研究工作和成果如下:

1. 本文提出了基于局部二值模式的改进时空上下文算法。通过对STC算法的实验发现,该算法存在由变形和遮挡引起的跟踪精度下降问题。针对上述问题,本文提出在原算法基础上引入局部二值模式和遮挡检测机制,利用LBP特征来代替灰度特征,当跟踪器检测出目标发生遮挡时,停止分类器参数的更新,利用目标先验信息对其进行位置预测以解决目标发生遮挡后的定位问题。经过试验分析,改进之后的算法能有效提升目标跟踪精度,针对遮挡目标也展现出良好的跟踪稳定性。

2. 本文提出了基于核相关滤波的层级遮挡检测算法。基于核相关滤波的快速跟踪算法在目标发生严重遮挡或全部遮挡时,该算法的定位精度会明显下降。经过试验观察,如果目标发生遮挡,跟踪器会错误地把背景信息当作目标信息,很可能引起定位精度的下降甚至失败。因此在该算法的基础上引入遮挡检测机制,当跟踪器判断出目标发生遮挡时,停止分类器参数的更新;同时利用LBP特征进行层级遮挡检测可以有效地区分目标变形和遮挡,利用目标先验信息对其进行位置预测以解决目标发生遮挡后的定位问题。经过试验分析,提出的层级遮挡检测算法能有效检测出遮挡并进行相应处理,展现出良好的跟踪稳定性。

3. 基于难度的目标跟踪性能评估方法改进。目标在不同场景下跟踪难度各不相同,如何定量描述目标跟踪难度有利于对跟踪算法优劣做出评判。本文优化了目标跟踪性能评估指标,利用灰度共生矩阵信息定量描述背景复杂度,利用灰度直方图定量描述背景与目标的相似度,利用目标遮挡比例定量描述目标的遮挡情况,利用边缘比率定量描述目标在背景中的重要性等。经过试验分析,本文提出的目标跟踪难度评价指标具有良好的适应性。

Other Abstract

As an important part of machine vision technology, target tracking technology has attracted more and more attention and has been widely applied in military and civil fields. However, due to the complexity and variability of the tracking environment, such as illumination changes, target deformation, occlusion, similar target interference and so on, the tracker is easily lost or misplaced during the tracking process, which leads to the tracking failure, and the occlusion causes the tracking failure main reason.

This paper focuses on the research and improvement of moving object tracking algorithms under complex scenes. Based on Spatial-Temporal Context (STC) algorithm and Kernel Correlation Filters (KCF) algorithm, this paper studies systematically the technology of blocking interference and designs a video target tracking difficulty evaluation system. It analyzes the background factors and their own factors in the target motion, and quantitatively describes the difficulty of tracking the target in different frames of different video sequences. Aiming at the problem of low accuracy and no occlusion detection mechanism in complex scenes, an improved spatial-temporal context algorithm based on Local Binary Pattern (LBP) and a hierarchical occlusion detection algorithm based on kernel-dependent filtering are proposed, which not only effectively improves The target localization accuracy in complex scenes is robust to occlusion of the tracking process. The main research work and achievements of this paper are as follows:

1. An improved spatial-temporal context algorithm based on local binary pattern is proposed. Through the experiment of the STC algorithm, it is found that the algorithm has the reduction of tracking precision caused by deformation and occlusion. To address this problem, we propose an improved method which adopts occlusion-detection strategy and uses Local Binary Pattern (LBP) to replace the gray feature. When the occlusion is detected by the tracker, the updates of classifier’s parameters are stopped.  For the object of linear motion, the objective prior information was used to predict the position of the object in order to solve the problem of occlusion. After the test analysis, the improved algorithm can effectively improve the tracking precision of the target, and also show good tracking stability for the occlusion target.

2. A hierarchical occlusion detection algorithm based on kernel correlation filtering is proposed. When the object is seriously or completely occluded, the tracking accuracy will decrease obviously. Through the analysis of KCF, we found that when the occlusion occurred, the classifier would introduce erroneous information, it is likely to cause a decline in positioning accuracy or even failure. Therefore, based on the KCF framework, this paper introduced the occlusion detection mechanism to stop the updating of the classifier parameters when the occlusion occurred,and using LBP feature for hierarchical occlusion detection can effectively distinguish target deformation and occlusion. As to the linear motion object, the objective prior information was used to predict the position of the object in order to solve the problem of occlusion. After the test analysis, the proposed hierarchical occlusion detection algorithm can effectively detect the occlusion and carry out the corresponding treatment, showing good tracking stability.

3. Video target tracking difficulty evaluation system design. It is difficult to track the targets in different frames of different video sequences. How to describe the target tracking quantitatively is helpful to judge the advantages and disadvantages of the tracking algorithms. In this paper, we propose a target tracking difficulty design scheme based on a variety of factors, using grayscale symbiotic matrix information to quantitatively describe the background complexity, using gray histogram to quantitatively describe the similarity between the background and the target, Proportions quantitatively describe the obstruction of the target, the use of edge ratio quantitative description of the importance of the target in the background. After experimental analysis, this article designed a video target tracking difficulty rating system has a good adaptability.

Subject Area图象处理
Language中文
Document Type学位论文
Identifierhttp://ir.ioe.ac.cn/handle/181551/8354
Collection光电技术研究所博硕士论文
Affiliation中国科学院光电技术研究所
Recommended Citation
GB/T 7714
崔盼果. 复杂场景下遮挡干扰目标鲁棒跟踪技术研究[D]. 北京. 中国科学院研究生院,2018.
Files in This Item:
File Name/Size DocType Version Access License
复杂场景下遮挡干扰目标鲁棒跟踪技术研究.(3977KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[崔盼果]'s Articles
Baidu academic
Similar articles in Baidu academic
[崔盼果]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[崔盼果]'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.