SIA OpenIR  > 机器人学研究室
基于多传感器融合的运动目标跟踪
其他题名Moving Object Tracking Based on Multi-sensor Fusion
黄微1,2
导师田建东
分类号TP391.41
关键词联合标定 图像配准 多传感器融合 目标检测与跟踪
索取号TP391.41/H78/2018
页数77页
学位专业模式识别与智能系统
学位名称硕士
2018-05-17
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门机器人学研究室
摘要随着机器视觉的发展,运动目标跟踪成为机器视觉研究领域的热点和难点,广泛应用在人机交互、虚拟现实、视频监控、机器人导航、智能汽车等领域,但在复杂多变的光照和气象条件下,视觉传感器的应用存在着一定的局限性,而多传感器融合可以实现信息互补,比单一传感器具有很好的容错性、实时性和鲁棒性,主要研究内容包括:首先,简要的介绍了多传感器配准融合算法的发展,详细的介绍了图像配准算法的基本理论及配准类型,并对多传感器配准融合算法的基本理论进行了论述。其次,介绍了毫米波雷达探测原理以及数据接收与解析方法,提出了一种毫米波雷达有效目标初选的方法,然后介绍了视觉传感器和毫米波雷达系统的联合标定方法,构建了毫米波雷达坐标系、世界坐标系、视觉传感器坐标系、图像物理坐标系和像素坐标系之间的转换关系,进而毫米波雷达和视觉传感器的空间配准转换关系。在此基础上,结合两个传感器的时间配准方法,完成两个传感器的时空配准。然后,研究了红外图像和可见光图像中的目标配准问题。在异源图像融合中,由于不同的拍摄条件、拍摄时间、成像原理等因素,配准方法的选择是一个难点。首先通过比较选择利用混合高斯模型对视觉传感器的光学图像与红外热像仪的红外图像进行前景检测,得到目标在图像序列中每一帧的形状轮廓信息;然后在此基础上,完成异源图像的特征点匹配、TPS形变模型转换、正则化和缩放特性处理;最后利用RANSAC随机抽样一致性算法去除误匹配,迭代得到光学图像与红外图像的图像匹配。最后,利用自适应背景相减法检测目标,将融合系统在实际环境中开展了运动目标跟踪模拟实验,验证了基于毫米波雷达和视觉传感器、红外热像仪和视觉传感器的感知系统的实际使用效果,分析了提出的基于多传感器融合的运动目标跟踪方法的可行性和有效性,为后续的系统改进提供了实验依据。
其他摘要With the development of machine vision, moving objects tracking has become a hot spot and a key in the field of this research. It is widely used in human-computer interaction, virtual reality, video surveillance, robot navigation, intelligent vehicles and so on, but under the complex and changeable light and weather conditions, the application of the visual sensor has some limitations, and the multi-sensor fusion can realize the information complementation. It has good fault tolerance, real-time and robustness compared to the single sensor. The main research contents are as follows. First of all, the development of multi-sensor registration and fusion algorithm is briefly introduced. The basic theory of image registration algorithm and the type of registration are introduced in detail. The basic theory of multi-sensor registration and fusion algorithm is elaborated. Secondly, the principle of millimeter-wave radar detection, data receiving and analysis methods are introduced, and a method for filtering out false objects by millimeter-wave radar is proposed. A joint calibration method for visual sensor and millimeter-wave radar system is introduced. The transformation relationships among world coordinate system, vision sensor coordinate system, image physical coordinate system and pixel coordinate system are constructed, spatial registration transformation relationship between millimeter-wave radar and visual sensor is obtained. On the basis of this, the space-time registration is completed with time registration method. Then, object registration in infrared images and visible images is studied. In heterogeneous image fusion, due to different shooting conditions, shooting time and imaging principle, registration method is difficult. The two images are pre-processed respectively by using the hybrid Gauss model to obtain shape contour of objects. Bipartite graph matching is established for the contour points of the two preprocessed images. In order to extend match to the whole shape, a TPS transformation model is established. A regularization method is used to relax the interpolation requirements, and the estimation error is reduced by iteratively reorganizing corresponding relationship and estimating the transformation. RANSAC algorithm is used to filter out wrong matching points, and make the tracking result more reliable. Finally, object detection is carried out by using the adaptive background subtraction method, fusion system is used to simulate moving object tracking in actual environment, and the practical application effect of the sensing system based on millimeter-wave radar and vision sensor, infrared thermal imager and camera is verified. The feasibility and effectiveness of the proposed moving object tracking method based on multi-sensor fusion are analyzed, which provide experimental basis for the subsequent system improvement.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/21811
专题机器人学研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
黄微. 基于多传感器融合的运动目标跟踪[D]. 沈阳. 中国科学院沈阳自动化研究所,2018.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
基于多传感器融合的运动目标跟踪.pdf(3084KB)学位论文 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[黄微]的文章
百度学术
百度学术中相似的文章
[黄微]的文章
必应学术
必应学术中相似的文章
[黄微]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

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。