Abstract
Assembling large-scale, high-definition maps from machine learning predictions generated on small, isolated geographic tiles can result in fragmented data, and some consolidation techniques may introduce geometric inaccuracies. A disclosed technique frames map consolidation as a global optimization problem. The process can involve partitioning a geographic area into patches and generating multiple prediction candidates for each patch. A global solver may then select a candidate for each patch to reduce a total cost. This cost function, for example, can evaluate the intrinsic quality of individual candidates and the geometric and semantic connectivity between candidates in adjacent patches. This approach can be used to assemble a large-scale, seamless, and topologically coherent map with improved geometric accuracy from disjointed model outputs.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Livne, Amir; Barzilai, Aviad; Veikherman, Danny; Emanuel, Dotan; Desheh, Yuval; Yair, Omer; Looi, Shiang Yong; Triess, Larissa; and Green, Rotem, "Global Optimization of Prediction Candidates for Scalable Map Consolidation", Technical Disclosure Commons, (July 20, 2026)
https://www.tdcommons.org/dpubs_series/11052