基于核相关滤波的运动目标跟踪算法研究
文献类型：学位论文
| 作者 | 谭舒昆 |
| 学位类别 | 硕士 |
| 答辩日期 | 2017-05-24 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 沈阳 |
| 导师 | 刘云鹏 |
| 关键词 | 运动目标跟踪 核相关滤波 高斯尺度空间 遮挡 支持向量机 |
| 其他题名 | Research of Moving Object Tracking Based on Kernelized Correlation Filters |
| 学位专业 | 控制工程 |
| 中文摘要 | 运动目标跟踪在视频处理方向有着广泛的应用场景，并且一直以来都是机器视觉领域一项重要的研究课题。虽然随着机器学习的发展，运动目标跟踪技术有了很大进展，但是受限于跟踪场景各种复杂因素的影响，要实现精确、快速、稳定的跟踪，仍然面临着巨大挑战。近年来，相关滤波的思想被应用到检测、跟踪领域，因其计算效率高且鲁棒性好，从而逐渐成为了学者们的研究热点。 本文以核相关滤波（Kernelized correlation Filter,KCF）目标跟踪算法为主要研究对象展开了深入研究，并且从两个方面对该算法进行了改进，主要研究内容及成果总结如下： （1）针对核相关滤波跟踪算法在跟踪过程中无法实现尺度自适应问题，提出了一种基于高斯尺度空间的相关滤波跟踪算法。即在核相关滤波跟踪算法的基础上结合高斯尺度空间，在进行跟踪的同时估计目标尺度变化。 （2）为了解决跟踪过程中目标丢失问题，即核相关滤波跟踪算法在跟踪过程中目标被大面积或完全遮挡后，无法在重新出现时被再次捕捉到，提出了一种基于支持向量机（SVM）的核相关滤波跟踪算法。该算法增加了遮挡处理机制，当目标被完全遮挡时使用支持向量机训练的再检测分类器进行检测，重新找到目标后恢复跟踪。 （3）将上述两种改进算法结合，提出了一种改进的核相关滤波跟踪算法。通过对实验结果的定性和定量分析表明，该算法与原算法及其他优秀跟踪算法相比，跟踪效果得到明显改善，跟踪精度有所提升。本文围绕当前目标跟踪领域的热点问题进行研究，并且有效解决了若干难点问题，可以为各种应用场景提供相应的理论基础。 |
| 英文摘要 | Moving object tracking has always been an important research topic in the field of machine vision, and has a wide range of application scenarios. Although with the development of machine learning, moving object tracking technology has made great progress, but limited by tracking the complex factors of the scene, to achieve accurate, fast and stable tracking, still face a great challenge. In recent years, the idea of correlation filters has been applied to the field of detection and tracking, which has become a hotspot because of its high computational efficiency and accurate tracking. In this paper, the Kernelized correlation filters object tracking algorithm as the main research object, mainly from two aspects of the algorithm to study and improve the main research content and results are as follows: (1) A correlation filters algorithm based on Gaussian scale space is proposed to solve the problem of scale adaptation in the tracking process. On the basis of the kernelized correlation filters algorithm, the Gaussian scale space is combined to estimate the target scale change while tracking. The results of experiment shows that the improved algorithm can achieve scale- adaptive. (2) In order to solve the problem of target loss in the tracking process, the kernelized correlation filters tracking algorithm can not be caught again when the target is large or completely occluded during the tracking process. A kernelized correlation filters algorithm based on support vector machine (SVM) is proposed. According to the occlusion processing mechanism, the re-detection classifier trained by the support vector machine is used to detect the target when it is completely blocked. The experimental results show that the method has a certain anti - occlusion effect. (3) An improved kernelized correlation filters algorithm is proposed by combining the two improved algorithms proposed earlier. The qualitative and quantitative analysis of the experiment shows that the algorithm is improved compared with the original algorithm and other excellent tracking algorithms. The effect is obvious, tracking accuracy has improved. This paper focuses on the hot issues in the field of tracking, and effectively solves a number of difficult problems, can provide a theoretical basis for a variety of application scenarios. |
| 语种 | 中文 |
| 产权排序 | 1 |
| 源URL | [http://ir.sia.cn/handle/173321/20540]  |
| 专题 | 沈阳自动化研究所_光电信息技术研究室 |
| 推荐引用方式 GB/T 7714 | 谭舒昆. 基于核相关滤波的运动目标跟踪算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所. 2017. |
入库方式： OAI收割
来源：沈阳自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。