Title:
MONOCULAR CAMERA-BASED THREE-DIMENSIONAL SCENE DENSE RECONSTRUCTION METHOD
Document Type and Number:
WIPO Patent Application WO/2019/174377
Kind Code:
A1
Abstract:
A monocular camera-based three-dimensional scene dense reconstruction method, belonging to the field of image processing and computer vision. The method comprises: modeling a camera pose and a three-dimensional scene by using a feature-based simultaneous localization and mapping method, deriving a fusion algorithm of traditional geometric depth estimation and convolutional neural network depth inference to establish a deep computational reconstruction model, and obtaining a high-quality dense depth map using a high-performance solution algorithm. The system is easy to be constructed, has low requirements for hardware resources, and can complete the dense reconstruction of a three-dimensional scene by means of a common monocular camera. The feature-based SLAM method is used to ensure the accuracy of localization, and fuses sparse depth scattering points and a CNN-inferred low-resolution depth map, enabling the recovery result to be more accurate. A fast and efficient algorithm is used to solve a deep reconstruction function, avoiding the problem of large-scale matrix inversion, and the algorithm is high in operation speed, thereby providing a guarantee for monocular camera-based real-time three-dimensional scene dense reconstruction.
Inventors:
YE XINCHEN (CN)
ZHONG WEI (CN)
WANG ZHIHUI (CN)
LI HAOJIE (CN)
LIN LIN (CN)
FAN XIN (CN)
LUO ZHONGXUAN (CN)
ZHONG WEI (CN)
WANG ZHIHUI (CN)
LI HAOJIE (CN)
LIN LIN (CN)
FAN XIN (CN)
LUO ZHONGXUAN (CN)
Application Number:
PCT/CN2019/070589
Publication Date:
September 19, 2019
Filing Date:
January 07, 2019
Export Citation:
Assignee:
UNIV DALIAN TECH (CN)
International Classes:
G06T17/00
Foreign References:
CN108416840A | 2018-08-17 | |||
CN106875482A | 2017-06-20 | |||
CN107194985A | 2017-09-22 | |||
CN106856012A | 2017-06-16 | |||
CN106157307A | 2016-11-23 |
Download PDF: