基于多傳感器融合的三維高斯?jié)姙R技術(shù)

打開文本圖片集

中圖分類號：TN911.73-34；TP242 文獻(xiàn)標(biāo)識碼：A 文章編號：1004-373X（2025）17-0093-05

引用格式：,.基于多傳感器融合的三維高斯?jié)姙R技術(shù)[J].現(xiàn)代電子技術(shù),2025,48(17)：93-97.

3DGaussian splatting technology based on multi-sensor fusion

LI Yongchang, LI Wei (Schoolof MechanicalandElectrical Engineering,China JiliangUniversity,Hangzhou31oo18,China)

Abstract：Tomeet thedemandof buildinghigh-quality3Dmodelsforintellgentmobilerobots,a3DGausiansplatting (3DGS)optimizationalgorithmbasedonmulti-sensorfusionisproposedbasedon3DGaussiansplattingreconstruction technology.ThedatasetcontainingLiDAR,cameraandIMUinformationisrecordedintherealscene.The3DGausian distributionisinitializedbytheoptimizedodometerposeandthecamerafeaturepointscalibratedbythelaserpointcloud.On thebasisoftakingaccountofthegeometricinformationoftepointcloud,thelossfunctionisimprovedfortheparameterodel optimization.The3DGSoptimizationalgorithmbasedonmulti-sensorfusionandtheoriginal methodweresubjectedto reconstructionexperimentsunderdiferentdatasets．Thereconstructedresultswerecomparedintworealscenarios.The experimentalresultsshowthatthepeak signal-to-noiseratio(SNR)ofthe3DGausiansplattingoptimizationalgorihmbasedon multi-sensor fusion isimproved by 3.5% and 3% in weak texture environment and general texture environment respectively,in comparisonwiththe3DGSreconstructionmethod.Anditsstructural similarity(SSIM)andlearnedperceptualimagepatch similarity(LPIPS)alsoperformwell.Itcanbeseenthatthequalityof thereconstructedmodelhasbeenimprovedand the effectiveness of the proposed algorithm has been validated.

Keywords：inteligentrobot;point cloud;3Dreconstruction;reconstructionaccuracy；3DGS;simultaneous localizationand mapping

0 引言

在現(xiàn)代機器人和自動駕駛技術(shù)中，SLAM（Simulta-neousLocalizationandMapping，即時定位與地圖構(gòu)建）系統(tǒng)扮演著至關(guān)重要的角色。（剩余11916字）