| 题名 | 基于特征匹配的机载电子稳像技术研究 |
| 作者 | 吉淑娇 |
| 学位类别 | 博士 |
| 答辩日期 | 2015-05 |
| 授予单位 | 中国科学院大学 |
| 导师 | 朱明 |
| 关键词 | 特征匹配 运动滤波 二进制算子 特征点分类 |
| 其他题名 | Airborne Electronic Image StabilizationTechnolofy Based on Feature Matching |
| 学位专业 | 机械电子工程 |
| 中文摘要 | 随着摄像系统在军用侦查系统，无人车辆导航系统以及航空测量系统，民用监控系统中的广泛应用，人们对所拍摄的视频序列的画面要求清晰稳定。然而由于上述诸多摄像设备或因所处工作环境恶劣，或因设备不稳定等会存在不确定的抖动和振动，使所采集的视频视觉效果变差，视频序列的质量有所下降，从而引起观察者的视觉疲劳，更增强了后续图像处理算法的难度。因此有必要采用电子稳像方法，去除或减少不规则的随机抖动，增强视频的质量。本论文以机载的航拍视频为研究对象，采用基于特征匹配的稳像技术对视频稳像。主要研究内容可归纳总结如下： 首先研究了电子稳像的背景、国内外发展现状以及机载稳像的关键技术和稳像评价方法；接着研究电子稳像系统的各主要环节包括运动估计、运动滤波以及运动补偿等环节的已有的算法，对各种算法进行分析比较；结合机载稳像系统的关键问题，提出了本文的创新算法，并通过实验对算法进行了有效性验证。 为了提高稳像过程运动矢量的计算速度，采用基于兴趣区的特征点匹配进行稳像。首先选取图像的兴趣区(ROI)并在其内进行特征点检测，接着将兴趣区内的特征点稀疏化，并在相邻帧的特征窗内利用SAD准则寻找匹配点对，最后利用LMeds准则剔除误匹配点并求解出全局运动矢量，再利用Kalman滤波及逐帧运动补偿。虽然算法采用基础传统的稳像算法，但是在诸多环节进行了改正和创新，比如特征点稀疏化过程，运动滤波部分等；通过实验结果，验证了算法的有效性和鲁棒性； 为了实时有效地去除机载成像设备所摄取视频序列的帧间抖动，采用基于二进制算子的特征匹配和加权最小二乘的粒子滤波的方法对视频序列进行稳像。首先利用BRISK算子提取相邻帧的稳定特征点，采用FREAK描述子对特征点进行描述并匹配，进而结合运动模型获取帧间粗运动矢量估计。最后引入粒子滤波解决因特征点景深不同引起的全局运动矢量计算不准确的问题，获得粒子权重后，采用加权最小二乘获得准确的全局运动矢量，该算法不仅提高了运动估计的速度，还有效地保证了运动估计的精度； 为了去除前景运动物体对稳像精度的影响，提出基于特征点分类的方法，将特征点分为前景特征点和背景特征点。首先通过KLT跟踪特征点，结合MSAC算法计算由初始运动估计所得匹配点和通过跟踪所得特征点之间的距离，然后根据阈值不断更新前景特征点集和背景特征点集，最后只采用背景上的特征点进行全局运动估计，有效的提高了含有复杂背景的运动视频的稳像效果。 |
| 英文摘要 | With the camera system widely utilized in military reconnaissance systems, unmanned vehicle navigation systems, aviation measurement systems and civilian monitoring system, people requires the video sequences captured from these camera system clear and stable. However, there will be jitter and vibration due to many of the above camera equipment are in bad working conditions or they were not stable, so that the visual effect of collected video is dropped, the quality of the video sequence has declined, this can easily cause the observer's visual fatigue, but also enhance the difficulty of subsequent image processing algorithms. Therefore it is necessary to use electronic image stabilization method to remove or reduce irregular random jitter, and enhance the quality of the video. In this paper, we take the airborne video sequences as the research object; try to stabilize the videos based on feature matching video stabilization technology. The main contents can be summarized as follows: First the background of the electronic image stabilization, the development situation at home and abroad as well as the key technologies of image stabilization and stabilization evaluation methods were studied in the paper, then the major components of electronic image stabilization systems include motion estimation, motion filter and motion compensation, analysis and comparison of variety existing algorithms; Combined with the key issues of airborne image stabilization system, innovative algorithms were proposed in this paper, experiments demonstrate the effectiveness of the algorithms. In order to improve the computing speed of the motion vector in the process of image stabilization, feature points (FPS) matching based on the region of interest area (ROI) is used to estimate motion vectors. First, we selected (ROI) of the input image, then the FPS were detected by Harris .Second to sparse the FPS and to find matching point in the adjacent frame use the rule of SAD inside the window. Finally, LMeds was applied to eliminate false matched points. Although the algorithm is based on the traditional image stabilization algorithm, the corrections and innovation methods were proposed in many segments, such as the process of features points sparseness, the part of motion filter, results show the effectiveness and robust of the algorithm. In order to remove the inter frame jitter of the video sequences recorded by airborne imaging equipment effectively, an algorithm of the combination of the feature matching based on BRISK operator and the weighted least square particle filter is proposed to stabilize the sequences. Firstly, the BRISK operator and the corresponding feature extraction model are introduced. Here the BRISK operator is employed to extract the feature points, the FREAK operator is utilize to describe the feature points and then combined with the motion model to obtain the inter frame motion vectors. Secondly, the particle filter with the weighted least-squares method is employed to solve the problem of the global motion vector calculation inaccuracy caused by different depth of the feature points. Finally, the Kalman filter is applied to separate the motion compensation components from the global motion vector for compensating the video sequences frame by frame. The theoretical analysis and simulation results indicate that the proposed algorithm not only guarantee the real-time of the image stabilization but also the accuracy of motion estimation. In order to remove the effects of moving foreground objects on the accuracy of image stabilization, feature points classification method is proposed in the paper, the feature points were divided into foreground feature points and background feature points. First, FPS were tracked by KLT, the distance between the FPS matched by original motion estimation and FPS obtained from the tracking by the algorithm MSAC was calculated, then foreground feature points set and background feature points set were updated according to the threshold value, finally global motion estimation was calculated by the FPS on the background only, it effectively improved the quality of stabilized video sequences with complex background motion. |
| 公开日期 | 2015-12-24 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ciomp.ac.cn/handle/181722/48844]  |
| 专题 | 长春光学精密机械与物理研究所_中科院长春光机所知识产出 |
| 推荐引用方式 GB/T 7714 | 吉淑娇. 基于特征匹配的机载电子稳像技术研究[D]. 中国科学院大学. 2015. |
| 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