Related Experiment Video
Updated: Jan 6, 2026
Digital Hybrid Model Preparation for Virtual Planning of Reconstructive Dentoalveolar Surgical Procedures
Published on: August 5, 2021
Design and validation of a low‑cost 3D intraoral scanner using structured‑light triangulation and deep‑learning
Ahmed M M Awad1, Ahmed Badway2, Lamiaa ElFadaly3
1Graduate student, Faculty of Engineering, Department of Mechatronics Systems Engineering, October University for Modern Sciences & Arts, Cairo, Egypt.
Statement Of Problem:
Intraoral scanners (IOSs) have transformed prosthodontic workflows by enabling precise, high-resolution digital scans. However, their high cost and hardware complexity limit adoption in resource-constrained settings.
Purpose:
The aim of this study was to design and validate a lightweight, cost-effective IOS prototype hardware using structured-light triangulation and deep-learning reconstruction and to compare its performance with a popular commercially available IOS (TRIOS 3).
Material And Methods:
A handheld prototype IOS hardware integrating a complementary metal-oxide-semiconductor (CMOS) camera (1280×720 px) with both white‑light and red‑laser projectors was developed. Intrinsic and extrinsic calibration used the Zhang method; feature extraction used Canny and scale-invariant feature transform (SIFT), structure‑from‑motion (SfM), and active triangulation generated point clouds in a photogrammetry software program. A YOLO‑V8-style network performed tooth segmentation, followed by a fully convolutional network (FCN) encoder-decoder for depth refinement. A gypsum cast was scanned (307 frames), and the 311 000 initial mesh points outputted were compared against the TRIOS 3 (102 000 points).
Results:
The mean ±standard deviation reprojection error of the prototype scanner hardware was 0.30 ±0.15 px (range 0.05 to 1.8 px), within commercial tolerances (0.2 to 0.4 px). The landmark count averaged 4000 ±1200 features per frame. After mesh filtering, 270 000 high‑quality vertices remained. Deep‑learning postprocessing reduced surface artifacts by approximately 20% (qualitative).
Conclusions:
The low‑cost IOS achieved point‑cloud densities 3 times higher than the commercially available IOS while maintaining comparable accuracy, demonstrating its potential in affordable digital prosthetic workflows. Future in vivo validation is planned to determine clinical applicability.
