以多数语义物体为主特征的语义地图重定位研究  

Semantic map relocalization using the primary features of most semantic objects

作　　者：蒋林[1,2] 明祥宇 汤勃 万乐 向贤宝 雷斌 郭宇飞 JIANG Lin;MING Xiangyu;TANG Bo;WAN Le;XIANG Xianbao;LEI Bin;GUO Yufei(Key Laboratory of Ministry of Education on Metallurgical Equipment and Control,Wuhan University of Science and Technology,Wuhan 430081,China;Institute of Robotics and Intelligent Systems,Wuhan University of Science and Technology,Wuhan 430081,China;Intelligent Research and Development Division,Wuhan Lianyi Heli Technology Co.Ltd,Wuhan 430081,China)

机构地区：[1]武汉科技大学冶金装备及其控制教育部重点实验室,湖北武汉430081 [2]武汉科技大学机器人与智能系统研究院,湖北武汉430081 [3]武汉联一合立技术有限公司智能研发部,湖北武汉430076

出　　处：《哈尔滨工程大学学报》2025年第2期363-373,共11页Journal of Harbin Engineering University

基　　金：国家自然科学基金项目(51874217);国家重点研发计划(2019YFB1310000);湖北省重点研发计划(2020BAB098)。

摘　　要：针对自适应蒙特卡罗定位算法在相似环境和长走廊环境及环境改变后定位不准的问题,本文提出一种以多数语义物体为主特征的语义地图重定位算法进行全局定位。该算法利用构建好的二维栅格语义地图提取语义物体的主特征,结合相机观测模型及主语义物体与周围次语义物体信息表实现全局预定位。通过预定位的结果改进粒子权重更新方式,提高自适应蒙特卡罗定位算法的实时性。结果表明:本文算法在室内相似环境及环境改变后定位速率较自适应蒙特卡罗定位算法分别提升了68.75%和52.78%,在长走廊环境及环境改变后定位速率较自适应蒙特卡罗定位算法分别提升了65.96%和53.13%,通过实验验证了本文算法在粒子收敛速率、鲁棒性、实时性都有提升。A semantic map relocalization algorithm with the majority of semantic objects as primary features for global localization is proposed to address the localization inaccuracies of the adaptive Monte Carlo localization(AMCL)algorithm in similar environments,long corridors,and after environmental changes.The algorithm initially extracts the primary features of semantic objects from a preconstructed 2D grid semantic map,then combines these features with the camera observation model and information table about the primary semantic objects and surrounding secondary semantic objects to achieve global prelocalization.Then,the particle weight update method is improved on the basis of the prelocalization results,ultimately enhancing the real-time performance of the AMCL algorithm.The results show that,compared with the AMCL algorithm,the proposed algorithm has improved the localization rate by 68.75%in similar indoor environments and by 52.78%after environmental change.In long corridor environments,after environmental changes,the localization rate has improved by 65.96%and 53.13%compared with that using the AMCL algorithm.Experiments confirm that the proposed algorithm can improve particle convergence rate,robustness,and real-time performance.

关 键 词：语义地图 主特征 相机 信息表 全局预定位 粒子 自适应蒙特卡罗定位算法 定位速率 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置]