基于多通道差融合的三维室内语义场景补全方法

打开文本图片集

关键词：多通道差融合；三维；室内；语义场景补全；RGB；Depth中图分类号：TP391 文献标识码：A文章编号：0439-8114（2026）02-0195-07DOI：10.14088/j.cnki.issn0439-8114.2026.02.029开放科学（资源服务）标识码（OSID）：口

3Dindoor semanticscene completion method based on multi-channel difference fusion

WANGChang-shuan1，LUYun-he²，JIANGJian-wu² （1.Guangxi Institute of SurveyingMapping and Geoinformation，Liuzhou 545oo6，Guangxi，China; 2.CollgeofGeomaticsandGeoinformation，Guilin UniversityofTechnology，Guilin541oo6，Guangxi，China）

Abstract：Toadesstheissuesofmissing3Dperceptalinformaionandisuffcientsemanticunderstandingcusedbybectocclusionandcompact spatial structures incomplex indoorenvironments，amulti-channeldiference fusionnetwork forsemanticsene completion（MCDFNet）basedonRGB-Dinput was proposed.Themodel designedamulti-channeldiferencefusion（MCDF）module，which，bsedoifiedGB-DepresetationetracteddierentalaturesmongGB，Depthandtheirfusedelsto ffectivelyenhancethemodelingcapabilityforthegeometricstructureandsemanticconsistencyofoccudedregions.Experimentson the NYUCAD dataset showed that the MCDFNet model achieved an accuracy of 72.8% ，aprecision of 77.1% ，and a mean Intersection overUnion （ mIoU ）of 43.4% while maintaining a single-scene completion inference time of 1.9 s，outperforming mainstream models such as AICNet，DDRNet，and GRFNet.Ablationstudies demonstrated thatintroducingthe MCDFmodulecould improve the mIoU by 1.5percentagepoits，provingitscriticalroleinnhancingcompletionaccuracy.Themodelcouldoperatestablyinighlyoudedindoorenvironments，improvingthecompletenessandpracticalvalueof3Dmaps，andwassuitableforvarious typicalindooraplcation scenarios.

Key Words：multi-channel diference fusion；3D;indoor；semantic scene completion；RGB；Depth

室内环境具有空间结构复杂、布局多样、受人类活动十扰显著及隐私保护要求高等特点，给机器人带来了严峻的空间感知与视觉感知挑战。（剩余10271字）