基于云的服务机器人语义地图构建研究

发布时间：2018-05-26 07:52

本文选题：语义地图 + 语义-CVFH样本库　；参考：《山东大学》2017年硕士论文

【摘要】：环境地图的语义描述程度是影响机器人智能化的关键要素,复杂环境中的"自主"语义感知能力是机器人导航智能化的重要体现。课题设计基于云的环境语义获取框架,确定按照类别划分的样本库方案、利用云端资源与样本库扩展本体知识库并实现语义地图的构建,将地图对物品、区域等的认知应用到机器人服务任务中来。针对机器人在服务任务中难以"自主"获得复杂环境中物品语义问题,设计基于云的环境语义获取框架。"云"虽然可以为我们提供海量资源,但这些资源并不能直接用于获取物品语义,需要进一步形成适用于获取语义信息的数据模型。框架基于卷积神经网络、支持向量机和点云技术形成语义-CVFH(Cluster Viewpoint-Feature Histogram)样本库,用于多样化场景的物品语义查询。为了提高导航精度并使地图更加人性化,将获取语义后的物品分为标识物品与归属物品(包括显性归属物品与隐性归属物品),其中标识物品与显性归属物品通过查询语义-CVFH样本库获得物品语义并分别存入标识库、归属库。标识归属物品位置关系通过对PCL(Point Cloud Library)点云进行处理确定。-复杂环境包含多种多样的物品,物品与环境的知识化是机器人实现智能化的重要因素。建立人、物品与房间为基础的本体模型,定义实例属性、对象属性并创建实例,利用云端资源、规则完善模型并持久化存储形成本体知识库。标识物品、归属物品与房间功能确定后用于本体知识库的发育,使机器人具备一定的推理能力,拥有类似于人的"常识"。在语义-CVFH样本库与本体知识库的基础上,机器人根据结构化环境信息(包括二维栅格地图与自定位),在物理坐标处分割场景点云得到物品点云并上传至私有云,通过查询语义-CVFH样本库获得物品语义;在结构化地图当前位置关联标识物品并通过XML标记语言记录此位置的标识归属位置关系形成语义地图,机器人根据需求由私有云下载语义地图用于执行服务任务。本文所提出的方法将人与机器人工作环境紧密关联,云端的丰富环境数据与机器人地图紧密关联,使机器人地图拟人化与知识化,为机器人提供智能化任务奠定了基础。该工作对深化和完善云机器人领域研究、加速其发展应用具有重要科学意义和实用价值。
[Abstract]:The semantic description of the environment map is the key factor affecting the robot intelligence. The "autonomous" semantic perception ability in complex environment is an important embodiment of the intelligent robot navigation. The subject design is based on the framework of the cloud environment semantics, determines the sample database scheme divided according to the category, and uses the cloud resources and the sample library to extend the ontology. The knowledge base and the construction of semantic map are applied to the task of robot service. It is difficult to obtain the semantic problems of the objects in the complex environment by the robot in the service task. Resources can not be used directly to obtain item semantics. It needs to further form a data model suitable for obtaining semantic information. The framework is based on convolution neural network, support vector machine and point cloud technology to form a semantic -CVFH (Cluster Viewpoint-Feature Histogram) sample library for multi sample semantic query of objects. In order to improve navigation essence, The objects and belongings (including the dominant belongings and the hidden belongings) are divided into the identification items and the belongings (including the dominant belongings and the hidden belongings), in which the items and the dominant belongings are obtained by the query semantic -CVFH sample library to obtain the object semantics and be stored separately in the library. The PCL (Point Cloud Library) point cloud is processed and determined. - complex environment contains a variety of objects. The knowledge of goods and environment is an important factor in the intelligentization of robots. Establish the ontology model based on people, objects and rooms, define instance attributes, object and create instances, use cloud resources, and improve the model with rules. And persisted to form the ontology knowledge base. Identifying objects, belongings and room functions are used for the development of ontology knowledge base, so that the robot has certain reasoning ability and has similar "common sense". Based on the semantic -CVFH sample library and ontology knowledge base, the robot is based on the structured environment information (including two-dimensional grid). Map and self positioning), divide the scene point cloud at the physical coordinates to get the object point cloud and upload it to the private cloud, and obtain the object semantics by querying the semantic -CVFH sample library. The semantic map is formed for the identification of the objects in the current position of the structured map and the identification of the location by the XML markup language. In order to download the semantic map from the private cloud, the semantic map is used to carry out the service task. The method proposed in this paper is closely related to the working environment of the robot. The rich environmental data in the cloud is closely related to the robot map, which makes the robot map personified and knowledgeable, and lays the foundation for the robot to provide intelligent tasks. The work is deepened and perfected. It is of great scientific significance and practical value to accelerate the development and application of cloud robots.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】