Healthy Research Rewards
ResearchHub is incentivizing healthy research behavior. At this time, first authors of open access papers are eligible for rewards. Visit the publications tab to view your eligible publications.
Got it
YL
Yixuan Li
Author with expertise in Graphene: Properties, Synthesis, and Applications
Achievements
This user has not unlocked any achievements yet.
Key Stats
Upvotes received:
0
Publications:
2
(0% Open Access)
Cited by:
0
h-index:
0
/
i10-index:
0
Reputation
Biology
< 1%
Chemistry
< 1%
Economics
< 1%
Show more
How is this calculated?
Publications
0

Fast fringe image denoising for 3D reconstruction using lightweight convolutional neural network

Linyue Zhou et al.Dec 4, 2024
Fringe projection profilometry (FPP) is widely used in the field of three-dimensional (3D) measurement, owing to its advantages of simple hardware configuration, high accuracy and speed. The traditional fringe projection systems mainly use visible light as a projection source, but they are not suitable for the measurement of human faces and shadow objects. Whereas infrared micro electro mechanical system (MEMS) project invisible infrared light and is more suitable for use as an illumination system for 3D measurement. However, the infrared fringe pattern generated by the MEMS projector usually carries a large amount of speckle noise, which leads to a limitation in the quality of 3D reconstruction. Therefore, in this paper, a lightweight convolutional neural network framework is proposed to achieve fast and high-precision denoising of infrared fringe images. A three-step phase shift method and a multi-view stereo phase matching method are used to compute the absolute phase for the denoised fringe images, achieving fast high-precision 3D reconstruction. Experiments are conducted in static and dynamic scenes, verifying that the lightweight network can improve the denoising speed while guaranteeing similar accuracy to conventional methods, while the designed system is capable of fast high-precision 3D reconstruction at a speed of more than 25 frames per second and an accuracy of 80µm.