Scene reconstruction of monocular endoscope video is essential for the enhancement of Surgical Endoscope Image
Analysis and Application. However, restricted by the narrow space of endoscopic movement and the obstruction of
vision within cavities, it’s difficult for most conventional methods to conduct high-quality reconstruction. To
overcome these challenges, a novel dynamic growing 3D gaussian splatting architecture is proposed to construct
the 3D model of endoscopic scene without pre-compute camera poses or Structure from Motion. Firstly, to establish
spatial feature associations between interframe, a 2D 3D displacement field is designed based on dense feature
matches and depth prediction. On this basis, a novel displacement field variational optimization is developed to
acquire relative poses by minimizing the energy functional of field transformation. Secondly, to address the
constraint of endoscopic view, by gaussian dynamic transformation and differential gradient field optimization,
a novel dynamic gaussian growing strategy is proposed to sequentially grow the local gaussian model. Finally, a
novel Forward-Reconstruction&Backward-Optimization architecture is proposed to generate the global gaussian model.
The evaluation is conducted on two public endoscopic datasets: Scared and C3VD. The experimental results show the
proposed method outperforms state-of-the-art methods in quantitative (PSRN, SSIM and LIPIS) and qualitative
comparisons. The project page is https://iheckzza.github.io/DG-3DGS/.
