Open AccessMonocular gray code 3D shape measurement based on improved Siamese Network
Author(s) -
Yangming Li,
Yan Piao,
Yuheng Liu
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1848/1/012162
Subject(s) - monocular , computer science , artificial intelligence , computer vision , code (set theory) , monocular vision , observational error , transformation (genetics) , system of measurement , set (abstract data type) , mathematics , biochemistry , statistics , chemistry , physics , astronomy , gene , programming language
As 3D information is widely used in more and more industries, simplification of 3D measurement operations and higher accuracy of measurement results are important development directions in the future. Compared with the binocular measurement system, simple equipment and convenient operation are the advantages of the monocular measurement; the disadvantage is that the measurement accuracy is low and it is greatly affected by the environment. This paper proposes a method that combines deep learning with traditional monocular Gray code 3D shape measurement technology. The space transformation network is used to improve the Siamese Network, and the single-view photos are taken through spatial transformation to obtain dual-view photos, and then matched through the Siamese Network. The purpose is to use monocular measurement equipment and methods to obtain the effect of binocular measurement. The deep learning network in this article has been pre-trained with a self-made data set before use. Compared with the original method, this method has higher accuracy in plane measurement, and has a significant improvement in depth measurement, successfully reducing the depth measurement error to less than 0.1mm.