基于语义嵌入三维高斯溅射的双路径重建算法OA

3DGS lacks the capability to incorporate semantic information within its explicit Gaussian distributions,which re-stricts its use in semantic perception reconstruction.In response to the above,this work proposed a dual-path reconstruction al-gorithm that embedded semantics into 3D Gaussian splatting.Firstly,the algorithm operated by projecting 2D semantic features from pre-trained models into 3D Gaussian distributions,utilizing their spatial relationships.Then,a dual-path collaborative ren-dering strategy fused fast rasterized rendering and depth-aware volume rendering to enhance output quality.Finally,an expo-sure-compensated image loss optimization model suppressed artifacts caused by lighting variations.Experimental results on the Tanks&Tempels dataset show that SDP-GS improves PSNR and SSIM by 9％and 8％,respectively,and reduces LPIPS by 41％compared to the baseline PSNR and SSIM.SDP-GS not only successfully embeds the semantic features into the 3D Gaussian splatting model,but also realizes the high-fidelity reconstruction.