A Highly Robust Automatic 3D Reconstruction System Based on Integrated Optimization by Point Line Features
Junyi Hou,Lei Yu,Shumin Fei
DOI: https://doi.org/10.1016/j.engappai.2020.103879
IF: 8
2020-01-01
Engineering Applications of Artificial Intelligence
Abstract:Current reconstruction systems often face the challenge of drifting when reconstructing complex scenes. Recent 3D(three-dimensional) reconstruction systems have shown convincing results, but still suffer from the following problems: (1) When the current vision-based 3D reconstruction system uses a single camera , the small angle of view of the camera is likely to cause the reconstructed 3D model to be incomplete. (2) Some image frames have fewer image feature points and image blurring, which leads to a larger deviation of the estimated camera pose value. (3) The current mainstream line feature 3D reconstruction system causes linearization and limits the update efficiency due to the adoption of the filter frame. In order to solve the above problems, this paper proposes a highly robust automatic 3D reconstruction system based on integrated optimization by point line features. Firstly, a multi-depth camera collaborative scanning method is developed to obtain a relatively complete 3D model. Secondly, a more accurate camera pose initial value can be obtained in advance without the position estimation. Thirdly, a comprehensive optimization method based on point line feature is used, which can improve the accuracy of camera pose and the consistency and accuracy of map construction. Many experiments show that the system can solve the problems of small viewing angle, blurred image and low modeling efficiency. The proposed system can be applied to 3D reconstruction of various complex large scenes. The obtained high-precision 3D model can be widely applied in the fields of human–computer interaction, virtual reality, etc.