空间语义地图机器人自主在线构建方法研究

Online autonomous building of spatial semantic map is currently a hot topic of mobie robot research. This project is designed to break through the limitations of existing research, and towards two semantic categories, namely object and place, to estabilish a novel method of online autonomous building of spatial semantic map for mobile robot. The topics forms three explicit and compatible contents: ①Semantic enviromental feature oriented online spatial map buidling, which includes objects composite feature extraction by using Kinect, improved SRUKF local SLAM method, and online spectrum partition of navigation map base on panoramic vision, will solve the efficiency issue of underlying SLAM and coordination between recognition system; ②Online learning and recognition of spatial environmental features for robot,which covers the construction method of Web-based object database, SVMs online incremental training and place recognition, will impove the adaptive ability of robot classifier; ③Quantization updating scheme of spatial semantic map, which contains map probabilistic representation and its cross-level propagation, proposing a probabilistic multiple hypothesis anchoring method for dynamic updating semantic map. The projects adopts Kinect and Omni-directioal vision as the main external sensors, and PeopleBot robot as the experimental platform to verify the relevant work. The research will greatly improve the advanced recognition level of robot, and lay a solid theoretical foundation for the practical application.

空间语义地图在线自主构建是移动机器人研究的新热点。本课题旨在突破既有研究的局限,面向物体及地点两种语义类别,建立空间语义地图的机器人自主在线构建的新方法。课题形成三个功能明确,兼容良好的研究内容:①面向语义环境特征的空间地图在线构建,研究基于Kinect物体复合特征提取,改进SRUKF局部SLAM,基于全景视觉的地图在线谱分割技术,解决底层SLAM效能和与识别系统的协作问题；②语义环境特征的机器人在线学习及识别,研究Web数据库模型库构建方法,SVMs在线增量训练方法以及地点识别方法,解决分类器在线学习适应性问题；③空间语义地图在线量化更新机制,提出概率量化表述及跨层次概率传播方法,提出概率多假设锚固PMHA解决语义地图动态量化更新问题。课题采用全景摄像机及Kineect传感器,以PeopleBot机器人验证相关工作。研究成果将极大提高机器人高级环境认知水平，为其实际应用奠定理论基础。

空间语义地图在线自主构建是移动机器人研究的新热点。本课题面向物体及地点两种语义类别,建立空间语义地图的机器人自主在线构建的新方法。课题形成三个主要研究内容:①研究物体特征提取方法及跟踪技术,提出机器人地图谱分割技术提高底层SLAM执行效率问题；②机器人地点类别识别方法；③空间语义地图在线量化更新机制,解决概率语义地图的动态更新问题。课题针对全景摄像机, Kinect,激光测距仪等,在实验室服务机器人上进行了相关验证工作。本项目共发表学术论文13篇，其中SCI检索论文4篇，EI检索论文11篇，中文核心期刊1篇，申请发明专利5项，授权3项，申请授权软件著作权1项，获得黑龙江省科技发明1等奖1项，培养博士生2名（已毕业），硕士研究生4名（已毕业），研究生3名（在读）。本项目为机器人空间认知开辟了道路，提升和推动了机器人空间感知及智能认知研究。