【作者】 王俊；

【导师】 姜欣；

【作者基本信息】 哈尔滨工业大学 ， 机械工程（专业学位）， 2019， 硕士

【摘要】 在房屋内部装修现场使用机器人有利于施工过程的标准化和系统化。随着劳动力成本的提高和机器人自动化技术的快速发展,越来越多的机器人用在自动化室内装修将成为必然趋势。然而如今机器人技术在该领域的应用仍然有限,自动化的室内装修还远未实现。造成这种状态的原因之一是现有的移动机器人同时定位与建图(SLAM,Simultaneous Localization and Mapping)的技术在环境背景单调和变化的室内装修现场难以有效,应用于自动化室内装修场景的机器人定位与环境感知技术尚未取得突破。在本文中,我们研究了一种专门用于机器人室内装修工作场景的SLAM方法并进行了相关实验。针对室内装修场景中环境背景单调,施工过程墙面视觉特征发生变化造成现有的应用于移动机器人的视觉SLAM方法鲁棒性差,精度低甚至无法使用的情况。我们提出并设计一种通过人工投影特征的应用在自主装修机器人上的视觉SLAM系统。该方案通过点阵激光投射器增加室内装修环境下的视觉特征,装修机器人使用ZED双目立体相机获取环境信息并提取视觉特征匹配并三角化建立局部地图,通过跟踪局部地图实现机器人的定位,在后端使用最小化重投影误差及闭环全局位姿矫正实现位姿及地图的优化,还通过对关键帧的点云进行拼接建立了环境的稠密地图。在我们现有的装修机器人上进行了测试,该方法实现了机器人在真实的室内装修环境的定位与地图构建。为了实现装修机器人室内装修工作过程的自动化,机器人需要定位并记录到待处理的物体如墙壁裂缝和缺陷点。为此我们提出并设计了融合墙面缺陷检测的装修机器人语义SLAM系统。该系统用深度学习的方法通过训练过的墙面缺陷检测网络从图像中识别出墙面缺陷,然后将缺陷的边界范围和种类信息融合到SLAM的前端,并结合SLAM前端估计出的帧间位姿对检测到的缺陷对象进行关联,最后利用缺陷边界框内特征点对应的地图点重建出的墙面缺陷并注册到SLAM地图中建立可供装修机器人自动化施工及导航用的语义地图。通过在真实的室内装修环境中进行试验,证明了该系统实现了准确识别墙面的缺陷并将其记录到用于机器人导航所构建地图中的功能,为装修机器人的自主化施工打下基础。更多还原

【Abstract】 The use of robots in the interior decoration of the house is conducive to the standardization and systemization of the construction process.With the increase of labor cost and the rapid development of robotic automation technology,more and more robots will be used in automated interior decoration.However,the application of robotics in this field is still limited,and the automation of interior decoration is far from being realized.One of the reasons for this situation is that the existing technology of SLAM(Simultaneous Localization and Mapping)is difficult to be effective in the interior decoration scene with monotonous and changing environment background,and the robot localization and environmental awareness Technology applied in the scene of automatic indoor decoration has not made a breakthrough.In this paper,we have studied a method of SLAM,which is specially used in the indoor decoration of robots,and carry out related experiments.In view of the monotonous environment background in indoor decoration scene and the change of wall visual characteristics during construction,the existing visual SLAM method applied to mobile robots has poor robustness,low accuracy and even can not be used.We propose and design a visual SLAM system for autonomous finsihing robots based on artificial projection features.The scheme adds visual features in indoor decoration environment by dot-matrix laser projector.The decoration robot uses ZED stereo camera to capture environmental information and extract visual features,match them,triangulate to build local map,and obtains the localization of the robot by tracking the local map.At the back end,the re-projection error and closure are minimized.The global pose correction of ring realizes the optimization of pose and map.Then the robot builds a dense map of the environment registering the point clouds of keyframes.Tested in our existing decoration robot,the autonomous robot has the ability of localization and mapping in the real interior decoration environment.In order to realize the automation of the interior decoration process of the finishing robot,the robot needs to obtain the position of the objects to be processed,such as cracks and defects in the wall.Therefore,we proposed and designed a visual SLAM system that integrates wall defect detection.The system uses the deep learning method to identify the wall defects from the image by training the wall inspection network,and then fuses the information of boundary range and type of the wall defects to the front-end of the SLAM,and correlate the detected defect objects according to the estimated inter-frame position and posture,and finally use the feature points in the defect boundary frame to correspond to each other.The wall defects reconstructed from map points are registered in SLAM map to establish a semantic map for automatic construction navigation of finishing robots.Through the test of the decoration robot in the real indoor decoration environment,the system realizes the function of accurately identifying the defects of the wall and registering them into the map built for the navigation of the robot,laying a foundation for the autonomous construction of the finishing robot.更多还原