MEMS激光3D视觉系统图像处理

发布时间：2018-04-20 23:08

本文选题：激光3D视觉系统 + 坐标变换　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：激光3D视觉系统具有较高空间分辨率和不受地面杂波干扰等优点,即使是在夜晚和复杂环境下也能获取空间目标的三维信息,因此其在无人车,智能机器人等环境感知领域有着重要的应用价值。目前,主流的激光3D视觉系统一般通过扫描系统获取空间目标的二维坐标信息、通过激光测距系统获取空间目标的一维纵向距离信息,然后将空间目标二维信息和一维距离信息进行融合,形成空间目标的三维信息。本文针对MEMS激光3D视觉系统中空间信息建模、激光点云数据处理和激光点云三维图像重构等关键问题进行研究,最后利用MEMS激光3D视觉系统获取的实验数据进行验证,主要研究内容如下:(1)简要介绍MEMS激光3D视觉系统的工作原理。实验上测量MEMS扫描角与控制电压之间的关系,寻找高空间覆盖率的MEMS扫描轨迹。建立激光3D视觉系统与目标之间径向距离变化模型,模拟仿真不同视场不同探测距离目标三维图像畸变规律,研究图像校正方法。建立大地坐标系到激光3D视觉系统坐标系统变换模型,为利用GPS或北斗导航系统定位打下理论基础。(2)研究激光点云数据预处理方法。比较分析各种点云邻域搜索算法,选择kd-tree作为本文点云数据搜索算法,进行法向量估计。采用统计滤波方法对奇异点进行剔除。比较分析数据光滑滤波的方法,通过仿真优选出双边滤波方法。(3)研究激光点云三维图像重构算法。根据Delaunay三角重构原理和三角网格优化准则,并结合MEMS扫描的特点,采用基于三角网格生长法的直接三角剖分。并对滤波后的圆台进行了仿真(4)利用已有的MEMS扫描激光3D视觉实验系统,对短距离单一目标进行成像,并对扫描过程中(x,y)坐标和z坐标不一一对应的现象提出了有效的校准方案。对不同距离下的复杂目标进行成像,对获取的激光点云数据进行预处理和三维图像重构,验证本文算法的可行性。
[Abstract]:Laser 3D vision system has the advantages of high spatial resolution and not being disturbed by ground clutter. It can obtain 3D information of space targets even at night and in complex environment, so it can be used in unmanned vehicles. Intelligent robots and other fields of environmental awareness have important application value. At present, the mainstream laser 3D vision system generally obtains the two-dimensional coordinate information of the space target through the scanning system, and the one-dimensional longitudinal distance information of the space target through the laser ranging system. Then, the two-dimensional information and one-dimensional distance information are fused to form the three-dimensional information of the space target. In this paper, the key problems of spatial information modeling, laser point cloud data processing and laser point cloud 3D image reconstruction in MEMS laser 3D vision system are studied. Finally, the experimental data obtained by MEMS laser 3D vision system are used to verify the problem. The main research contents are as follows: 1) briefly introduce the working principle of MEMS laser 3D vision system. The relationship between MEMS scan angle and control voltage is measured experimentally to find the MEMS scan track with high space coverage. A model of radial distance variation between laser 3D vision system and target is established, and the distortion law of 3D image from different field of view is simulated, and the method of image correction is studied. The transformation model from geodetic coordinate system to laser 3D vision system is established, which lays a theoretical foundation for using GPS or Beidou navigation system to study the preprocessing method of laser point cloud data. By comparing and analyzing various point cloud neighborhood search algorithms, kd-tree is selected as the point cloud data search algorithm in this paper to estimate the normal vector. The statistical filtering method is used to eliminate the singularity. Comparing and analyzing the methods of data smoothing filtering, the algorithm of laser point cloud 3D image reconstruction is studied by selecting the two-sided filtering method by simulation. According to the principle of Delaunay triangulation and the optimization criterion of triangulation mesh, combined with the characteristics of MEMS scanning, a direct triangulation method based on triangular mesh growth method is adopted. The filtered platform is simulated and simulated. Using the existing MEMS scanning laser 3D vision experimental system, the short range single target is imaged. An effective calibration scheme for the phenomenon that the coordinates and z coordinates do not correspond to each other in the scanning process is put forward. The complex target at different distances is imaged, the laser point cloud data is preprocessed and the 3D image is reconstructed, and the feasibility of this algorithm is verified.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】