Changchun Institute of Optics,Fine Mechanics and Physics,CAS
Data-Driven Neural Network Model for Robust Reconstruction of Automobile Casting | |
Lin, J.; Y. Wang; X. Li and L. Wang | |
2017 | |
发表期刊 | 3D Research |
卷号 | 8期号:3 |
摘要 | In computer vision system, it is a challenging task to robustly reconstruct complex 3D geometries of automobile castings. However, 3D scanning data is usually interfered by noises, the scanning resolution is low, these effects normally lead to incomplete matching and drift phenomenon. In order to solve these problems, a data-driven local geometric learning model is proposed to achieve robust reconstruction of automobile casting. In order to relieve the interference of sensor noise and to be compatible with incomplete scanning data, a 3D convolution neural network is established to match the local geometric features of automobile casting. The proposed neural network combines the geometric feature representation with the correlation metric function to robustly match the local correspondence. We use the truncated distance field(TDF) around the key point to represent the 3D surface of casting geometry, so that the model can be directly embedded into the 3D space to learn the geometric feature representation; Finally, the training labels is automatically generated for depth learning based on the existing RGB-D reconstruction algorithm, which accesses to the same global key matching descriptor. The experimental results show that the matching accuracy of our network is 92.2% for automobile castings, the closed loop rate is about 74.0% when the matching tolerance threshold is 0.2. The matching descriptors performed well and retained 81.6% matching accuracy at 95% closed loop. For the sparse geometric castings with initial matching failure, the 3D matching object can be reconstructed robustly by training the key descriptors. Our method performs 3D reconstruction robustly for complex automobile castings. 2017, 3D Research Center, Kwangwoon University and Springer-Verlag GmbH Germany. |
收录类别 | sci |
语种 | 英语 |
文献类型 | 期刊论文 |
条目标识符 | http://ir.ciomp.ac.cn/handle/181722/59052 |
专题 | 中科院长春光机所知识产出 |
推荐引用方式 GB/T 7714 | Lin, J.,Y. Wang,X. Li and L. Wang. Data-Driven Neural Network Model for Robust Reconstruction of Automobile Casting[J]. 3D Research,2017,8(3). |
APA | Lin, J.,Y. Wang,&X. Li and L. Wang.(2017).Data-Driven Neural Network Model for Robust Reconstruction of Automobile Casting.3D Research,8(3). |
MLA | Lin, J.,et al."Data-Driven Neural Network Model for Robust Reconstruction of Automobile Casting".3D Research 8.3(2017). |
条目包含的文件 | 下载所有文件 | |||||
文件名称/大小 | 文献类型 | 版本类型 | 开放类型 | 使用许可 | ||
Data-Driven Neural N(3067KB) | 期刊论文 | 作者接受稿 | 开放获取 | CC BY-NC-SA | 浏览 下载 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论