复杂环境下面向激光扫描点云的三维目标定位方法研究

As one of the key techniques in envirionment perception for robot system, three-dimensional (3D) object accurate localization is a challenging issue in the field of robot. For traditional charge coupled device (CCD) image-based object localization methods, it is difficult to realize accurate localization of 3D objects in complex environment with continuously varying illumination, objects and scenes since they are strongly affected by other factors, such as light conditions and shadows. Therefore, this project will thoroughly study the problems of 3D object localization by using laser scanning point clouds of complex environment. We aim to design a set of object feature extraction and localization algorithms for robot complex environment, and these algorithms have the advantages of high precision and strong robustness. Based on accurate estimation of the measurement model parameters of 3D laser scanners, this project will convert the problems, namely the feature extraction and localization of 3D objects of point clouds, to robust optimization problems by using scanning point clouds of robot complex working environment. This project will solve these optimization problems by using swarm intelligence algorithms such as particle swarm optimization (PSO) and invasive weed optimization (IWO). And the scene features can be extracted robustly, and the 3D object pose parameters can be precisely estimated. The project will provide reliable technical methods and theoretical foundation for robot to accurately perceive the complex environment. Meanwhile, all the related research achievements can be employed in many robot domains, such as environment perception, object grasping, path planning, autonomous navigation and target tracking.

作为机器人系统环境感知中的一项关键技术，三维目标精确定位是机器人领域一项极富挑战的课题。受光照和阴影等因素影响，传统基于CCD图像的目标定位方法难以实现光照、目标和场景不断变化的复杂环境下三维目标的精确定位。本项目旨在利用复杂场景的激光扫描点云数据，对复杂环境下三维目标定位问题展开深入研究，进而设计出一套面向机器人复杂环境的目标特征提取和定位算法，且具有高精度、强鲁棒的优点。本项目在精确估计三维激光扫描仪测量模型参数的基础上，利用机器人复杂工作场景的扫描点云，拟将点云中三维目标特征提取和定位问题转化为鲁棒非线性优化问题，采用粒子群优化和入侵性杂草优化等智能优化算法进行优化求解，以鲁棒提取场景特征信息并精确估计三维目标位姿。本项目的实施可为机器人对复杂环境的精确感知提供可靠的技术手段和理论基础，研究成果可应用于环境感知、物体抓取、路径规划、自主导航和目标跟踪等机器人领域。

三维目标精确定位作为机器人系统环境感知的关键技术之一，是机器人领域一项极富挑战的课题。本项目利用智能机器人复杂环境的激光扫描点云数据，对复杂环境下三维目标精确定位问题展开研究，其主要研究内容包括：（1）三维激光扫描仪标定算法研究；（2）点云特征提取方法研究；（3）面向点云的三维目标定位方法研究。本项目取得的重要结果包括：（1）提出的基于伪Huber损失函数的三维激光扫描仪精确标定算法，能够有效抑制测量噪声和外点对标定结果影响；（2）提出了基于多维粒子群优化的三维点云特征提取方法，并将其推广应用于无监督图像分割中；（3）提出的基于生成树的点云配准方法，能够有效克服噪声影响，提高多视图点云的配准精度，并提出了基于点云配准的三维目标精确定位方法。. 本项目所研究的复杂环境下面向激光扫描点云的三维目标定位方法，对于提高机器人系统对复杂环境的感知和适应能力，具有重要的理论和现实意义。本项目成果在环境感知、物体抓取、路径规划、自主导航和目标跟踪等机器人领域具有广泛的应用前景。