Abstract: | Aiming at the issues of low confidence and significant error in the maps generated by traditional cloud fusion methods for high-definition maps, a crowd-sourced high-definition map cloud fusion method based on factor graph optimization was proposed. Using RTK-GPS data, the local semantic map uploaded by the vehicle was processed globally. Then the map fragments were matched and the matching accuracy was improved by the consistency screening process. Subsequently, lane matching was employed to construct a factor graph optimization model for the transformation relation matrix between constraint maps. Finally, the real multi-vehicle trajectory data from a city road was used for testing. The results show that, compared with the traditional methods, the proposed algorithm has a lower dependence on initial optimization values, and improves the aggregation degree between maps by 44.7%. Compared with the actual lane data values, the absolute errors of the proposed algorithm are consistently within 1m, indicating high practical value. |