AI-Based Single-Shot Structured-Light Depth Reconstruction for Real-Time Laparoscopic Surgical Guidance
Original title:AI-based single-shot structured-light depth reconstruction for real-time laparoscopic surgical guidance
This paper presents a synchronization-free, single-shot structured-light depth system for laparoscopic guidance. A passive LED-illuminated binary mask is paired with a dual-channel laparoscope, while a VQ-VAE provides discrete latent representations and a custom U-Net predicts depth directly without a separate mask branch. Using 722 paired phantom images with Zivid-referenced depth, the method reports 3.70 mm MAE, 0.0326 AbsRel, delta=1.1 accuracy of 0.962, and delta=1.1^2 accuracy of 0.970. It processed 301 consecutive frames at 26.0 Hz on an NVIDIA A100 GPU.
Why it's worth reading
The work addresses a concrete integration bottleneck by replacing synchronized multi-shot projection with single-shot video-rate sensing, while its phantom-only evaluation makes the validation gap important for surgical robotics.