| 题名 | 基于粒子滤波的小目标检测与跟踪方法研究 |
| 作者 | 赵永廷 |
| 学位类别 | 硕士 |
| 答辩日期 | 2012-05-28 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 中国科学院沈阳自动化研究所 |
| 导师 | 肖阳辉 |
| 关键词 | 小目标检测 背景抑制 局部自适应阈值 粒子滤波 状态估计 |
| 其他题名 | Research on the method of small target detection and tracking based on the particle filter |
| 学位专业 | 控制理论与控制工程 |
| 中文摘要 | 小目标是指图像平面中所占像元数少、信噪比低的目标。小目标的检测与跟踪技术广泛地应用于军事与民用领域，对小目标检测与跟踪算法进行深入研究，具有重要的意义。但是由于小目标所固有的特性，使得小目标检测与跟踪技术难度大，其主要体现在以下几个方面：信噪比低；亮度和位置存在不连续性； 缺少形状、纹理等目标特征；缺少背景先验统计信息等。目前国内外众多学者和研究机构都在致力于提高小目标检测算法的性能，并结合具体应用，开展了大量的研究工作，小目标检测成为一项研究热点。 本文围绕提高小目标检测与跟踪的性能，在“基于滤波和数据关联”目标跟踪框架下，提出了一种基于粒子滤波的小目标检测跟踪方法。该方法主要通过提高单帧图像中小目标检测精度和融合序列图像间的目标状态信息的方式，解决了对亮度不连续并且机动性较高的小目标的稳定跟踪问题。 单帧图像小目标检测主要包含两个环节：目标信息的增强和目标的检出。通过对目前现有的背景抑制方法原理的分析和实验对比，结合本文的具体应用情况，通过NWTH背景抑制方法实现了小目标图像信噪比的提升；针对小目标亮度不连续的特点，本文提出了一种基于局部自适应阈值分割的小目标检出方法。通过大量的实验，验证了方法的有效性。 在帧间信息的融合方面，本文提出了一种基于粒子滤波的状态估计方法。由于粒子滤波算法对非线性非高斯过程具有较强的鲁棒性，可以对机动性较强的非线性过程实现最优逼近。为了提高状态估计结果的精度，本文详细的分析研究了运动过程的数学建模问题，对各类模型在粒子滤波算法下的具体实现，给出了分析推导过程。对于当前状态模型噪声分配系数的确定，文中提出了合理的假设，基于该假设，确定了具体的参数。经过仿真实验，验证了该算法的有效性。 本文实现了单帧图像中小目标的检出方法与序列图像间基于粒子滤波的目标状态估计方法的有机结合， 完成了基于粒子滤波的小目标检测与跟踪方法。通过实验分析验证了最终算法的合理性和有效性。最后对本文进行了总结，提出了小目标检测的研究展望。 |
| 索取号 | TN911.73/Z48/2012 |
| 英文摘要 | The small target is composed with few pixels, and it owns low signal noise rate. The relative research about the small target detection and tracking is used in many areas, including military and civil fields. There are many difficult elements influencing on the target detection. For example, lack statistic information about background, extreme lowest signal and noise rate, instability of the small target, etc. So scholars and institutes at home and abroad contribute on the research of small target detection in order to improve the detection performance. Recently it has become a hot and difficult topic. In order to improve the performance of small target detection, the framework of “data filter and association “is adopted, a method of small target detection and tracking based on particle filter is proposed. Depending on improving the performance of detesting the target in single image and merging the information, non-continues problem and the problem of target maneuvering is solved. There are two important parts about the target detection in single image, enhance the target in the image and detect target. The information of background is suppressed as noise, so the research on background suppression is done. According to theoretical analysis and experimental proves, the most effective method is choose. Target detection is very important part, and a target detection based on local adaptive threshold is proposed. Consider kinds of element influencing the target detection, this method own adaptability. In the factor of merging information, a state estimation based on particle filter is proposed. Particle filter is robust to non-linear and non-Gaussian process, so that adopting this algorithm can get optimal estimation of maneuvering target. Mathematical model of target motion is researched in order to describe the motion accurately. The problem to use kinds of the Mathematical model in particle filter is resolved. The noise distribution coefficient is determined under reasonable assumptions. Finally, two important parts are integrated together, and a method of small target detection and tracking is proposed. Experiments are done to prove this method. Paper is summarized and further research is prospected. |
| 语种 | 中文 |
| 产权排序 | 1 |
| 公开日期 | 2012-07-27 |
| 分类号 | TN911.73 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.sia.ac.cn/handle/173321/9236]  |
| 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 | 赵永廷. 基于粒子滤波的小目标检测与跟踪方法研究[D]. 中国科学院沈阳自动化研究所. 中国科学院沈阳自动化研究所. 2012. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
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。
修改评论