Affiliation:
1. Xidian University
2. Beijing Institute of Space Mechanics & Electricity
Abstract
The shape from polarization is a noncontact 3D imaging method that shows great potential, but its application is limited by the monocular camera system and surface integration algorithm. This Letter proposes a novel, to the best of our knowledge, method that employs deep neural networks to enhance multi-target 3D reconstruction, making a significant advancement in the field. By constructing the relationship between targets’ blur, distance, and clarity, the proposed method provides accurate spatial information while mitigating inaccuracies arising from the continuous model. Experiments show that the constructed neural network can help improve the multi-target 3D reconstruction quality compared with conventional methods.
Funder
National Natural Science Foundation of China
China Postdoctoral Science Foundation
CAS Key Laboratory of Space Precision Measurement Technology
Subject
Atomic and Molecular Physics, and Optics