Research on static denial scene reconstruction and target recognition technology based on 3DGS

XinYu Wu; Yin Zhu; YingSheng Wang; ZhaoYang Zhang

doi:10.63313/JCSFT.2004

Authors

XinYu Wu North Automatic Control Technology Institute, Taiyuan 030006, China Author
Yin Zhu North Automatic Control Technology Institute, Taiyuan 030006, China Author
YingSheng Wang North Automatic Control Technology Institute, Taiyuan 030006, China Author
ZhaoYang Zhang North Automatic Control Technology Institute, Taiyuan 030006, China Author

DOI:

https://doi.org/10.63313/JCSFT.2004

Keywords:

Denial scene, 3D Gaussian Splatting, Scene Reconstruction, Target Recognition

Abstract

Aiming at the problem of limited perception of unmanned equipment in denial scenes, this paper proposes a technical framework of scene reconstruction and target recognition based on 3DGS (3D Gaussian Splatting), which realizes efficient modeling and accurate representation of static denial scenes. Specifically, the semantic attributes are extended in the 3D Gaussian primitives and the feature index mechanism is introduced to improve the semantic modeling ability of 3DGS in the case of low memory occupation, and the target recognition is carried out on this basis. Experiments show that the framework can effectively meet the actual functional requirements of scene reconstruction and target recognition for unmanned equipment in static denial scenarios.

References

[1] MILDENHALL B, SRINIBANSAN P, TANCIK M, et al. Nerf: Representing scenes as neural radiance fields for view synthesis [J]. Communications of the ACM, 2021, 65(1): 99 -106.

[2] ZHANG J, LIU X, YE X, et al. Editable free- viewpoint video using a layered neural r- epresentation[J]. ACM Transactions on G- raphics(TOG), 2021, 40(4): 1-18.

[3] PARK K, SINHA U, BARRO J T, et al. Nerfies: Deformable neural radiance fields [C]// Proceedings of the IEEE/CVF internation- al conference on computer vision. 2021: 58 65-5874.

[4] PUMAROLA A, CORONA E, PONS-MOLL G, et al. D- nerf: Neural radiance fields for dynamicl scenes [C]//Proceedings of the IEEE/CVF conference on computer vision and patte- rn recognition. 2021: 10318-10327.

[5] DU Y, ZHANG Y, YU H X, et al. Neural radian- ce flow for 4d view synthesis and video processing[C]//2021 IEEE/CVF Internati- onal Conference on Computer Vision (ICC V). IEEE Computer Society, 2021: 14304-14 314.

[6] YANG Z, YANG H, PAN Z, et al. Real-time pho- torealistic dynamic scene representati- on and rendering with 4d gaussian splat- ting[J]. arXiv preprint arXiv:2310.10642, 2023.

[7] WU G, YI T, FANG J, et al. 4d gaussian splat- ting for real-time dynamic scene render- ing[C]//Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. 2024: 20310-20320.

[8] KIM M, LIM J, HAN B. 4d gaussian splatting in the wild with uncertainty-aware regu- larization[J]. Advances in Neural Infor- mation Processing Systems, 2024, 37: 12920 9-129226.

[9] YANG Z, PAN Z, ZHU X, et al. 4d gaussian spl- atting: Modeling dynamic scenes with na- tive 4d primitives[J]. arXiv preprint ar- Xiv:2412.20720, 2024.

[10] YANG L, ZHU K, TIAN J, et al. WideRange4D: Enabling High-Quality 4D Reconstruction with Wide-Range Movements and Scenes[J]. arXiv preprint arXiv:2503.13435, 2025.