【作者】 刘鹏；

【导师】 王春阳；

【作者基本信息】 长春理工大学 ， 控制工程（专业学位）， 2019， 硕士

【摘要】 三维重构是三维智能信息处理的核心技术,在数字城市、智能交通、逆向工程、文物修复、医学图像处理等领域有着广阔的应用背景。传统激光点云三维重构主要是基于PC平台完成,不仅效率低下,而且其重量体积功耗等均无法满足当前无人机载成像激光雷达系统等场合的应用。随着电子技术发展,多核DSP的应用在很多场合取代了PC机的功能,在主频、功耗、体积、运算能力等方面较PC机均具有突出优势,并具备强大的并行处理能力,非常适合大规模点云处理。因此,开展基于多核DSP的三维并行重构技术研究,不仅能显著提高点云处理及表面重构的效率,同时还能提升成像系统灵活性,对三维实时重构技术发展及应用具有非常重要的意义。鉴于此,本文以激光点云为研究对象,对三维重构中去噪、配准及表面重构等关键技术的并行处理展开了深入研究,主要研究内容包括以下三个方面:1)基于多核DSP平台的点云离群点并行剔除技术实现。深入研究了基于k-means聚类的离群噪点剔除算法,并在基于PC平台的MATLAB软件中进行算法仿真实验,验证了其可行性;提出了一种基于多核DSP平台的点云离群点并行剔除方案,采用了主从模式任务架构,利用IPC MessageQ实现了核间通信,并在基于多核DSP平台的CCS软件中进行了仿真实验,最后与基于PC平台的处理结果做了对比分析。2)基于多核DSP平台的点云并行配准技术实现。深入研究了点云配准过程中特征点检测、特征描述子计算、特征点匹配等关键技术,重点研究了基于改进特征点匹配的迭代最近点(ICP)配准算法,并在PC平台进行了算法仿真实验,验证了其可行性;提出了一种基于多核DSP的点云并行配准方案,并基于多核DSP平台进行了点云并行配准仿真实验,最后与基于PC平台的配准结果做了对比分析。3)基于多核DSP平台的点云表面并行重构技术实现。深入研究了基于生长法的Delaunay三角网格重构算法,并在PC平台进行了算法仿真实验,验证了算法的可行性;提出了一种基于多核DSP的点云模型并行重构方案,并基于多核DSP平台进行了点云并行重构仿真实验,最后与基于PC平台的重构结果做了对比分析。更多还原

【Abstract】 3D reconstruction is the core technology of three-dimensional intelligent information processing,which has a broad application background in the fields of digital city,intelligent transportation,reverse engineering,cultural relics restoration and medical image processing and others.The traditional 3D reconstruction of laser point cloud is mainly based on PC platform,which is not only inefficient,but also its weight and volume cannot meet the current application of the UAV imaging LIDAR system and other occasions.With the development of electronic technology,the application of multi-core DSP has replaced the function of PC in many occasions.Compared with PC platform,it has more outstanding advantages in terms of dominant frequency,power consumption,volume and computing power,and has powerful parallel processing capability,which is very suitable for large-scale point cloud processing.Therefore,the research on 3D parallel reconstruction technology based on multi-core DSP platform can not only significantly improve the efficiency of point cloud data processing and surface reconstruction,but also enhance the flexibility of imaging system.which is of great significance for the development and application of 3D real-time reconstruction technology.In view of this,this paper takes the laser point cloud as the research object,and the parallel processing of key technologies such as de-noising,registration and surface reconstruction of point cloud in 3D reconstruction is in-depth researched.The main research contents include the following three aspects:1)A technology of the outliers of Point cloud parallel removed based on multi-core DSP platform is implemented.The algorithm of the outliers of Point cloud removed based on k-means clustering is deeply studied,and the simulation experiments of the algorithm was carried out in MATLAB software based on PC platform to verify its feasibility.A scheme of point cloud parallel removed based on multi-core DSP platform is proposed.which adopted master-slave mode to process tasks,realized inter-core communication by IPC MessageQ,and the simulation experiments in CCS software based on multi-core DSP platform was carried out.Finally,the results of simulation experiments were compared with those based on PC platform.2)A technology of parallel registration of point cloud based on multi-core DSP platform is implemented.The key techniques of the detection of feature point,the calculation of feature descriptor and the matching of the feature point in the process of point cloud registration were deeply studied.The ICP registration algorithm based on the improved matching of feature point is researched.The simulation experiments of the algorithm were carried out on PC platform to verify its feasibility.A scheme of the parallel registration of point cloud based on multi-core DSP is proposed.The simulation experiments of parallel registration of point cloud based on the multi-core DSP platform were carried out.Finally,the results of simulation experiments were compared with those based on PC platform.3)A technology of surface parallel reconstruction of point cloud based on multi-core DSP platform is implemented.The algorithm of Delaunay triangular mesh reconstruction based on growth method is deeply studied.The simulation experiments of algorithm were carried out on PC platform to verify the feasibility of the algorithm.A scheme of parallel reconstruction of point cloud based on multi-core DSP is proposed,and the simulation experiments of the parallel reconstruction of point cloud based on the multi-core DSP platform were carried out.Finally,the results of simulation experiments were compared with those based on PC platform.更多还原