基于单目视觉的港口AGV自主导航关键技术研究

发布时间：2018-02-12 09:41

本文关键词： 港口AGV 自主导航单目视觉里程计路径规划动态避障　出处：《集美大学》2015年硕士论文　论文类型：学位论文

【摘要】：我国港口集装箱吞吐总量已经连续十年位居全球第一,然而这却给传统的港口运输模式带来空前压力,国外在20世纪90年代开始将AGV系统应用于港口水平运输,从而大幅度提高了港口集装箱的运转效率,但目前应用于港口AGV的导引系统都是基于固定路径的模式,这种方式简单可靠,但灵活性较差,智能化程度不高。近年来,随着图像处理技术和计算机技术的发展,基于视觉导航技术受到广泛关注,本课题基于NI的Starter KIT移动小车搭建了单目视觉系统,利用单目视觉技术解决了港口AGV的自主定位、路径规划和避障等问题,对港口AGV完全自主导航的实现进行了探索性研究。主要工作如下:(1)港口AGV单目视觉里程计定位系统设计。利用摄像机获取的连续图像信息推算港口AGV的位姿变化,主要针对SIFT特征点提取和匹配进行了改进,提出了一种实时性和准确性都较好的单目视觉里程计算法,进而能够实时准确获取AGV的位姿。(2)港口AGV的全局路径规划。根据港口环境结构化程度较高的特点,考虑了AGV运行时能量消耗和安全性,对A*算法的评价函数进行了改进,设计了一种适合港口环境的全局路径规划算法。(3)港口AGV的局部路径规划。以单目视觉系统实时获取的道路平面为样本,基于道路平面的颜色特征提出了一种自适应图像分割的障碍物检测算法,利用单目几何定位技术确定障碍物的位置,根据子目标和障碍物的位置选择通往目的地的最佳航向角。(4)硬件设计和导航系统软件编程。搭建了港口AGV实验平台,利用D-link-DIR617无线路由器搭建局域网,实现了AGV、IP摄像机和上位机的无线通信,采用Lab VIEW和MATLAB混合编程设计了港口AGV自主导航和远程控制软件。(5)在搭建的模拟港口环境内完成了港口AGV实验车的远程控制和自主导航实验。在远程控制模式下,根据IP摄像机获取的实时道路信息,通过Lab VIEW发布的共享变量对AGV进行远程控制。在自主导航模式下,港口AGV实验车完成了自主定位、路径规划和避障实验,实验结果表明基于单目视觉系统的自主导航算法具有较好的鲁棒性。
[Abstract]:The total volume of container throughput in China has been ranked first in the world for ten consecutive years. However, this has brought unprecedented pressure to the traditional port transportation mode. In 1990s, foreign countries began to apply AGV system to port level transportation. Therefore, the efficiency of port container operation is greatly improved, but at present, the guidance system used in port AGV is based on the fixed path mode. This method is simple and reliable, but the flexibility is poor, and the degree of intelligence is not high in recent years. With the development of image processing technology and computer technology, visual navigation technology has been paid more and more attention. In this paper, a monocular vision system based on NI Starter KIT mobile vehicle is built, and the independent positioning of port AGV is solved by using monocular vision technology. Path planning and obstacle avoidance, This paper makes an exploratory study on the realization of complete autonomous navigation of port AGV. The main work is as follows: 1) the design of port AGV monocular visual mileometer positioning system. The position and orientation changes of port AGV are calculated by using the continuous image information obtained by the camera. In this paper, the feature points extraction and matching of SIFT are improved, and a monocular visual mileage calculation method with good real-time and accuracy is proposed. Furthermore, the global path planning of the port AGV can be obtained in real time and accurately. According to the characteristics of the higher structural degree of the port environment, the energy consumption and security of the AGV running time are considered, and the evaluation function of the A * algorithm is improved. In this paper, a global path planning algorithm for port environment is designed. The local path planning of port AGV is presented. The road plane obtained in real time by monocular vision system is taken as a sample. Based on the color features of road plane, an adaptive obstacle detection algorithm for image segmentation is proposed. The hardware design and software programming of navigation system are designed according to the optimal heading angle to the destination. The port AGV experimental platform is built, and the local area network is built by using D-link-DIR617 wireless router. Realized the wireless communication between the AGVN IP camera and the host computer. The port AGV autonomous navigation and remote control software. 5) is designed by using Lab VIEW and MATLAB. The remote control and autonomous navigation experiment of the port AGV experimental vehicle is completed in the simulated port environment. According to the real-time road information obtained by IP camera, the remote control of AGV is carried out through the shared variables published by Lab VIEW. In autonomous navigation mode, the port AGV experimental vehicle has completed autonomous positioning, path planning and obstacle avoidance experiments. Experimental results show that the autonomous navigation algorithm based on Monocular vision system is robust.
【学位授予单位】：集美大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U653.92;TP23

【参考文献】