基于深度摄像和ARM的AGV避障技术研究

发布时间：2018-07-01 16:52

本文选题：AGV避障 + 深度摄像　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：自动导航运输车是一类通过传感器探测地形,实现在物理空间中定向移动的自动机器人车辆。AGV避障为该领域的研究重点之一,其主要内容是在实际的生产环境中,AGV能够按照特定的避障策略,保证车辆能够无碰撞地从起点运行到终点。AGV避障方法中的一个核心技术就是障碍物识别:探测和分析空间环境,定位环境中的障碍物信息。环境探测是AGV进行障碍物识别的前提条件,机器人可以通过计算机视觉来获取实时环境数据。本论文采用Kienct作为主传感器获取深度图,再运用梯度检测和DBSCAN聚类算法分析环境中的障碍物,然后将分类后的障碍信息映射到栅格图中,最后采用D*寻路算法完成AGV的路径规划。本论文的主要内容为:1.根据AGV车辆生产环境的特殊性,采用ARM架构的树莓派作为AGV的主控制器。由于Kinect摄像头缺乏在ARM平台上的设备驱动,先自主编写完成ARM平台上的Kinect设备驱动。2.引入梯度检测算法作为地平面检测手段,结合深度图梯度检测算法和DBSCAN聚类算法,快速有效地解析深度数据。首先阐述了Kienct和其生成深度图的原理,再提出了深度图像的梯度概念,以地平面和物体间的三维空间关系作为参考信息,运用梯度检测算法快速区分可行驶区域和障碍。3.设计了基于Kinect深度摄像头和ARM架构树莓派主控制器的AGV系统的整体框架。AGV利用Kienct探测运行环境,获取物体分布数据生成深度图。并利用经梯度检测和聚类算法处理后的深度图,生成描述AGV与障碍间位置关系的栅格图,最后运用启发式的D*路径规划方法的辅助下,进行AGV的动态路径规划。
[Abstract]:Automatic navigation vehicle is a kind of automatic robot vehicle. AGV, which detects terrain through sensors and realizes directional movement in physical space, is one of the key research points in this field. The main content is that in the actual production environment, AGV can follow a specific obstacle avoidance strategy, and one of the core technologies in the collision free vehicle running from the starting point to the end. AGV obstacle avoidance method is obstacle identification: detecting and analyzing space environment. Locate obstacle information in the environment. Environment detection is the precondition of obstacle recognition for AGV. Robot can acquire real-time environment data by computer vision. In this paper, Kienct is used as the main sensor to obtain the depth map, then gradient detection and DBSCAN clustering algorithm are used to analyze the obstacles in the environment, and then the classified obstacle information is mapped to the grid map. Finally, the path planning of AGV is completed by using the D * path finding algorithm. The main content of this thesis is 1: 1. According to the particularity of AGV vehicle production environment, raspberry pie based on arm is used as the main controller of AGV. Because Kinect camera lacks the device driver on arm platform, we write the Kinect device driver on arm platform. 2. The gradient detection algorithm is introduced as a ground plane detection method, combining with the depth map gradient detection algorithm and DBSCAN clustering algorithm, the depth data can be analyzed quickly and effectively. Firstly, the principle of Kienct and its depth map generation is expounded, then the concept of gradient of depth image is proposed. The 3D spatial relationship between ground plane and object is taken as the reference information, and the gradient detection algorithm is used to quickly distinguish the tradable area from the obstacle. 3. The whole framework of AGV system based on Kinect depth camera and arm raspberry master controller is designed. The AGV uses Kienct to detect the running environment to obtain the depth map of object distribution data. Using the depth map processed by gradient detection and clustering algorithm, the grid map describing the relationship between AGV and obstacle position is generated. Finally, the dynamic path planning of AGV is carried out with the aid of heuristic D* path planning method.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】