面向工业互联网的井下无人机单目视觉SLAM定位方法

发布时间：2018-06-27 08:54

本文选题：工业互联网 + 无人机　；参考：《北京交通大学》2017年硕士论文

【摘要】：通过井下工业互联网,可以对井下无人机等智能设备进行远程管理,对实现无人采矿具有十分重要的意义。为了对井下无人机进行管理,需要实现矿井环境下的无人机自主定位和导航。同时定位与建图(SLAM)算法可以通过传感器对周围环境信息进行观测,对无人机进行精确定位。因此,本文针对井下巷道中行驶的无人机单目视觉SLAM算法进行了研究,以实现矿井环境下的无人机自主定位和导航。本文的主要成果如下:(1)针对较宽阔无障碍物的巷道以及较窄巷道分别提出在巷道顶壁设置带有位置信息的二维码以及在巷道壁两侧设置反光标识牌作为无人机引导路标。针对这两种不同的巷道环境,分别创建了基于几何-拓扑的井下巷道图。(2)针对宽阔巷道环境,提出了一种含有位置信息的二维码,根据边缘检测、拟合直线等算法得到二维码的码值信息,即二维码的位置信息,实验和仿真表明该算法能快速清晰地对二维码进行识别。针对较窄巷道环境中设置的人工路标即反光标识牌,事先将每一个反光标识牌和其周围的自然特征进行图像获取并放入库中,提出了基于RANSAC的SIFT算法对井下无人机单目视觉所获取的每一帧图像进行特征提取,并与事先建立好的特征图像库进行匹配。实验和仿真表明基于RANSAC的SIFT算法有很高的正确匹配率,可以根据特征图像库和离线地图得到路标位置信息。(3)针对较宽阔无障碍物的巷道和二维码无人机引导路标场景,提出了一种基于二维码的井下无人机单目视觉PSOFastSLAM算法。仿真结果表明所提出的井下无人机PSOFastSLAM算法,有效改善了 FastSLAM定位算法粒子退化的问题,提高了井下无人机定位精度。(4)针对较窄巷道和反光标识牌引导路标场景,提出了一种基于反光标识牌的井下无人机单目视觉EKF-SLAM算法,通过已知路标得到的观测信息对无人机位姿进行估计。仿真结果表明单目视觉EKF-SLAM算法可以对井下无人机进行精确定位。最终结果显示,针对不同的巷道环境,采用不同的SLAM算法可以对面向工业互联网的井下无人机进行精确定位,为后续工业互联网对采集环境数据的井下无人机进行有效管理打下基础。
[Abstract]:Through the underground industrial Internet, intelligent equipment such as underground UAV can be managed remotely, which is of great significance to the realization of unmanned mining. In order to manage the underground UAV, it is necessary to realize the autonomous positioning and navigation of the UAV in the mine environment. The simultaneous location and Mapping (slam) algorithm can accurately locate the UAV by using sensors to observe the surrounding environment information. Therefore, in order to realize autonomous positioning and navigation of UAV in mine environment, the single vision slam algorithm of unmanned aerial vehicle (UAV) driving in underground roadway is studied in this paper. The main achievements of this paper are as follows: (1) for the wider roadway without obstacles and the narrower roadway, a two-dimension code with position information on the top wall of the roadway and a reflective sign on both sides of the roadway wall are put forward respectively as the UAV guide sign. Aiming at these two different laneway environments, the underground roadway map based on geometry and topology is created respectively. (2) aiming at the wide roadway environment, a kind of two-dimensional code with location information is proposed, which is based on edge detection. The code value information of the two dimensional code, i.e. the position information of the two dimensional code, is obtained by fitting the straight line. The experiment and simulation show that the algorithm can recognize the two dimensional code quickly and clearly. In view of the manual signpost set in the narrower tunnel environment, that is, the reflective sign, each reflective sign and its surrounding natural features are obtained and put into the library. This paper presents a sift algorithm based on RANSAC for feature extraction of each frame of image acquired by Monocular vision of downhole UAV, and matches it with the pre-established feature image database. Experiments and simulations show that the sift algorithm based on RANSAC has a high correct matching rate and can be used to obtain the location information of road signs according to the feature image database and off-line map. (3) aiming at the wide roadway without obstacles and the two-dimension code UAV guiding signpost scene, A PSOFastSLAM algorithm for downhole UAV monocular vision based on two dimensional code is proposed. The simulation results show that the proposed PSOFastSLAM algorithm can effectively improve the particle degradation of the FastSLAM algorithm and improve the positioning accuracy of the downhole UAV. (4) for the narrower roadway and the reflective sign to guide the road sign scene, the simulation results show that the proposed algorithm can effectively improve the particle degradation of the FastSLAM algorithm and improve the positioning accuracy of the downhole UAV. This paper presents an EKF-SLAM algorithm for downhole UAV monocular vision based on reflective sign, which can estimate the UAV position and attitude through the observation information obtained from the known road signs. Simulation results show that Monocular vision EKF-SLAM algorithm can accurately locate downhole UAV. The final results show that different slam algorithms can be used to locate the downhole UAV for industrial Internet in different laneway environments. It lays the foundation for the following industrial Internet to manage the underground UAV which collects environmental data effectively.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD67;V279

【参考文献】