Our paper entitled “Advancing Dense Endoscopic Reconstruction with Gaussian Splatting-driven Surface Normal-aware Tracking and Mapping” has been accepted at #ICRA2025! ๐
Multi-view inconsistencies in 3D Gaussian Splatting (3DGS) have long limited precise depth and surface reconstruction in minimally invasive surgery. Our team (Yiming Huang *, Beilei Cui *, Long Bai *, Zhen Chen, Jinlin Wu, Zhen Li, Hongbin Liu, Hongliang Ren) from The Chinese University of Hong Kong and Centre for Artificial Intelligence and Robotics, Hong Kong Institute of Science & Innovation, CAS introduces Endo-2DTAMโa real-time endoscopic SLAM system that combines 2DGS with a surface normal-aware pipeline to achieve geometrically accurate, high-quality reconstruction.
๐ Key Innovations:
โ Robust tracking via point-to-point/plane metrics
โ Surface enhancement using normal consistency & depth distortion
โ Pose-consistent keyframe sampling for coherence
๐ Results:
โ 1.87ยฑ0.63 mm RMSE in depth reconstruction
โ Real-time rendering & efficient computation
โ Open-source code: GitHub (https://lnkd.in/gyRr2YxS)
Paving the way for safer, smarter surgical robotics! ๐ค๐ก
