Abstract
Three-dimensional Gaussian splashing technology provides a new representation paradigm that combines high fidelity and computational efficiency for dynamic reconstruction and real-time interaction of facial expressions in virtual portraits. This article systematically discusses the mechanism of virtual portrait expression modeling and interactive implementation based on a 3D Gaussian model. At the theoretical level, the geometric mathematical representation, dynamic properties, and differentiable rendering support of Gaussian primitives were elucidated, and the modeling requirements for spatiotemporal continuity of facial expression motion were analyzed. At the methodological level, a mapping architecture from the expression parameter space to Gaussian attributes was constructed, and deformation fields based on temporal continuity, high-frequency micro-expression adaptive refinement strategies, and geometric texture fidelity constraints were proposed. At the interactive level, real-time inference paths, multimodal coupling frameworks, anti-drift mechanisms, and sparse computing collaborative optimization schemes were designed. Theoretical analysis and methodological verification indicate that this framework can effectively support low-latency and high-stability immersive virtual portrait interaction.
References
[1] Luo JJ, Tang SJ, Li SB, et al. Real-time 3D Gaussian semantic modeling method based on superpoint graph consistency. Bull Surv Mapp. 2026;(S1):230-235.
doi:10.13474/j.cnki.11-2246.2026.S142.
[2] Ji SZ, Wang DC, Li J, et al. Air-ground integrated 3D Gaussian reconstruction method based on implicit surface constraints. Bull Surv Mapp. 2026;(S1):106-111.
doi:10.13474/j.cnki.11-2246.2026.S118.
[3] Fu L, Zhu ZH, Li YQ. Research on 3D reconstruction of large public buildings in low-illumination indoor environments based on Gaussian splatting technology. J Inf Technol Civ Eng Archit. 2026;18(2):96-101. doi:10.16670/j.cnki.cn11-5823/tu.2026.02.15.
[4] Zhang ZL, Yang SM, Wang L, et al. Research on ore 3D color reconstruction method based on 3D Gaussian splatting. China Min Mag. 2026;35(4):105-113.
[5] Li XJ, Chen SY. Speech-driven 3D Gaussian facial motion generation method. High Technol Lett. 2026;36(3):221-229.
[6] Li H, Zhang J, Xia QL, et al. Research on fruit tree 3D visualization based on 3D Gaussian splatting. Agric Technol Equip. 2026; (3):1-4,7.
[7] Cheng L, Niu R, Zeng K, et al. Robot waypoint annotation method based on virtual scenes. Electr Eng. 2026;27(3):52-56,62.
[8] Liu YJ, Liu WC, Pan H, et al. An efficient lightweight method for feature-preserving real-time 3D rendering on the web. J Univ Chin Acad Sci. 2026;43(2):240-251.
[9] He J, Qian JJ, He L, et al. Research on tree reality-based 3D surveying and mapping method using 3D Gaussian splatting technology. Des Hydroelectr Power Stn. 2026;42(1):15-18.
doi:10.16671/j.cnki.cn51-1382/tv.2026.01.002.
[10] Song CS, Xu N, Yang RQ, et al. Dynamic scene 3D reconstruction for robotic surgery based on 3D Gaussian splatting. Acta Opt Sin. 2026;46(5):155-171.
[11] Cai S, Chen R, Ye XY, et al. Modeling and rendering of transmission towers based on improved 3D Gaussian splatting. Remote Sens Inf. 2026;41(1):70-78.
doi:10.20091/j.cnki.1000-3177.2026.01.009.
[12] Liu S, Hong M, Zhang J, et al. Modeling and robust control for power insurance claim collaborative robots using 3D Gaussian splash reconstruction. J Phys Conf Ser. 2025;3062(1):012005. doi:10.1088/1742-6596/3062/1/012005.