Semantic feature-based environment mapping and matching localization which has strong environmental adaptability is a research hotspot in the field of autonomous driving and robotics,and semantic map registration is one of the key links.Most of the existing registration methods only consider the simple quantitative and hierarchical distribution relationship between semantic features,which is prone to registration failure in highly repetitive semantic feature environments such as urban roads,resulting in larger map errors and decreased localization accuracy.Based on the idea of cliques in graph theory,a semantic map registration method that takes local structure into account is proposed.Firstly,based on the relative distance and orientation relationship be-tween semantic features with more significant differences,the attribute maps containing local structure information are constructed and the similarity calculation criterion between attribute maps is established.Secondly,a semantic map registration method for coarse-fine cascades is de-signed,and a robust initial value of the map registration is obtained by keeping local structure when matching attribute graphs in the coarse registration stage,and then the semantic feature point cloud is utilized for fine registration to improve the success rate of map registration.Based on the KAIST Urban dataset,the proposed algorithm is tested and validated,and compared with the classical algorithms such as neighbourhood graph(NGraph)and semantic histogram descriptor(SHD),the registration success rate of using the proposed algorithm is improved by about 93.5％and 74.0％,respectively,and map registration can be completed robustly and accurately without relying on prior poses,even in high similarity scenarios with semantic repetition.
NETL
NSTL
万方数据
