基于距离选通激光成像的三维目标重建与识别方法研究
文献类型：学位论文
| 作者 | 王明明
|
| 学位类别 | 博士 |
| 答辩日期 | 2015-11-30 |
| 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 | 朱枫 ; 郝颖明 |
| 关键词 | 激光成像 距离选通 三维重建 三维目标识别 |
| 其他题名 | Researches on Laser Range Gated Imaging Based Three Dimentional Object Reconstruction and Recognition |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 在目标探测识别领域，二维图像识别已经取得了较好的成果，二维图像反映的是物体的光谱反射特性，而物体还具有一定的三维几何形状，也可以用于目标识别，前提条件是需要三维传感器获取物体的三维信息。距离选通激光成像结合了非扫描激光主动成像技术和距离选通技术，通过激光器发射激光脉冲照射目标，由选通相机接收目标对激光脉冲的反射回波以实现目标成像同时获得目标距离信息，选通成像可以抑制后向散射噪声以提高图像质量。通过使用滑动门技术对物体不同距离进行多次成像，可以获取目标的距离选通切片图像序列及切片距离信息以实现物体三维成像，继而通过三维重建可以提供物体的三维几何信息用于目标识别。距离选通激光成像在目标探测识别领域具有广阔的应用前景，开展基于距离选通激光成像的三维目标重建与识别方法研究具有重要的理论意义和实际应用价值。关于距离选通激光成像的三维重建与识别目前存在以下两方面的问题：首先，距离选通序列切片数据与三维点云和网格数据存在显著差异，结合距离选通激光成像的特点建立序列切片数据向三维点云或网格数据的转换是三维目标识别的基础；其次，三维目标识别采用的特征描述符维数普遍较高，而且有时为了稳定性而牺牲描述性，使得三维目标的识别效率与精度有限。本论文将从上述两个问题着手，以实现基于距离选通序列切片数据的三维目标识别为目标，研究分析制约三维目标识别的关键问题；并通过建立仿真平台和实验原型完成对理论方法的实验验证。本文的主要研究内容如下：(1) 距离选通序列切片三维重建针对成像过程中由激光强度的空间分布特性引起的目标光照不均匀、由大气湍流引起的不同切片光照强弱不一致、由成像参数可能引起的目标表面不连续等问题，从强度像序列切片灰度值矫正、相邻切片数据插值等角度对切片数据进行了处理，建立了适用于三维目标识别的均匀网格表达。(2) 基于三维模型几何特征的目标识别针对现有的局部特征描述符Spin Image维数过高以及丢失形状信息等现象，从设计由低维特征和高维特征组成的新复合特征、调整匹配识别策略的角度对Spin Image方法进行了改进，并针对误匹配建立了快速可靠的正确特征匹配选择策略，以提升识别的效率和精度。 (3) 融合反射数据与距离数据的三维目标识别针对单纯的三维距离数据难以区分形状相同但反射属性不同物体的问题，通过将物体表面几何特征与反射特征相融合，同时根据反射特征与几何特征的维度差异建立了由粗到精的匹配策略以提升识别效率和精度。(4) 算法验证平台建立与算法验证在上述理论研究的基础上，首先根据距离选通激光成像原理建立了仿真系统以生成模拟距离选通序列切片数据用于重建与识别算法的初步验证，而后采集了距离选通成像的真实数据对理论方法进行实验验证，验证了理论方法的可行性。本文的研究工作对基于距离选通激光成像下的三维目标重建与识别进行了有益的尝试，有助于推动距离选通激光成像技术在实践中的应用。 |
| 英文摘要 | Two-dimensional (2D) images have been extensively used in targets detection and recognition with markable achievements. The 2D images reflect the spectral reflectance properties of an object, and the three-dimensional (3D) geometric information can also be useful for target recognition. Special 3D sensor is needed to obtain the 3D geometric information of an object. Laser range gated imaging system is a combination of non-scanning laser active imaging technology and range gated technology, laser emitter is used to generate laser pulse, range gated camera receives the laser reflection echo and generates a target image containing distance information that backscattering influence has been restrained. Based on the images generated using sliding gate technology, ranged-gated sequence slices image and slice’s distance information can be obtained that can further be used for 3D imaging of an object. 3D reconstruction can provide 3D geometric information for target detection. Therefore, researches on laser range gated imaging based 3D targets detection and recognition technology is of great theoretical significance and practical application value. There are two issues that retard the researches currently: firstly, the sequence slices data obtained by the laser range gated imaging are different from point cloud or mesh, it should be converted properly for the target recognition; secondly, the descriptors used for 3D target recognition are always with high dimension that makes the recognition less efficient. In this paper, we will address these two issues, analyze the critical problems encountered, establish the methods for 3D reconstruction and recognition, and verify the method by simulation and experimental test. The main research contents are as follow: (1) Laser range gated images based 3D reconstruction Aiming at the problem of the uneven spatial distribution of laser intensity, the inconsistent intensity of laser imaging, and the surface discontinuous, we try to establish a more proper mesh description for 3D target recognition by the intensity image gray value correction, adjacent slices data interpolation. (2) 3D model geometry characteristic based object recognition Aiming at the problem of Spin Image descriptor’s high dimension and shape information lossing, we try to establish a new 3D geometric feature descriptor, which consists of a low dimensional feature and a high dimensional feature, and adjust the recognition matching strategy, also build a fast and reliable strategy for correct matching selection, to enhance the efficiency and accuracy of target recognition. (3) Reflected image and depth image based 3D object recognition Aiming at the problem that the 3D distance data cannot be used to distinguish the targets with the same shape but different reflection property, we try to combine the geometrical features of the object with the reflection feature, establish matching strategy according to the dimension difference between the reflection features and geometrical features, to enhance the efficiency the efficiency and accuracy of target recognition. (4) Algorithm verification platform building and algorithm verification On the basis of the above theoretical researches, the simulation system has built up for the laser range gated imaging to generate slices data for the initial verification of the theoretical methods, then the real data of the laser range gated imaging is used for further verification. The research work of this paper is a useful attempt to the 3D object reconstruction and recognition based on the laser range gated imaging system; it is helpful to promote the application of the laser range gated imaging system in the practice. |
| 语种 | 中文 |
| 公开日期 | 2015-12-27 |
| 产权排序 | 1 |
| 页码 | 109页 |
| 源URL | [http://ir.sia.ac.cn/handle/173321/17524]  |
| 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 | 王明明. 基于距离选通激光成像的三维目标重建与识别方法研究[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. 2015. |
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。