Roadside　monitoring　Lidars　(RMLs)　will　be　a　crucial　part　of　the　future　intelligent　transportation　system.　Current　approaches　for　optimizing　RMLs＇　placement　at　intersections　work　in　hypothetical　environments　which　do　not　well　reflect　real-world　situations.　This　article　proposes　a　new　virtual　method　(VM)　for　optimizing　the　deployment　of　RMLs　at　as-built　intersections.　The　proposed　VM　operates　in　a　virtual　environment　where　both　static　background　and　dynamic　agents　are　modeled　by　dense　point　clouds.　The　agents　are　driven　by　real-world　motion　data.　Using　RMLs＇　parameters　as　inputs,　a　coarse-to-fine　subsampling　approach　is　developed　to　generate　laser　scans　in　the　virtual　world.　An　objective　function　is　then　defined　by　comparing　the　agents＇　points　in　the　generated　laser　scan　sequences　against　their　original　models.　Bayesian　optimization　is　applied　to　maximize　the　objective　function　by　setting　the　RMLs＇　positions　and　poses　as　decision　variables.　Besides,　batch　processing　strategy　and　parallel　computing　are　used　to　accelerate　the　optimization　process.　The　effectiveness　of　the　proposed　VM　is　demonstrated　in　a　case　study.　The　VM　shall　help　road　administrators　make　decisions　on　RMLs＇　deployment　at　as-built　intersections.