Publication Type : Book Chapter
Publisher : Springer
Source : In Advanced Machine Intelligence and Signal Processing, pp. 845-853. Springer, Singapore, 2022.
Url : https://link.springer.com/chapter/10.1007/978-981-19-0840-8_65
Campus : Coimbatore
School : School of Engineering
Department : Center for Computational Engineering and Networking (CEN)
Year : 2022
Abstract : Deep neural network is widely used nowadays for the extraction of geometrical features from the 3D models. Extracting the geometrical features from the 3D bodies plays an important role in many applications like registration and tracking. This paper focuses on the registration of CAD models of machining features which are commonly used in industries. Data augmentation is done in order to obtain the pairwise dataset. For the registration, our approach obtained a pairwise dataset of the CAD models and used metric learning to train the fully convolutional geometrical features. The resulting model is able to obtain pairwise matching points during registration. As evaluation step, feature retrieval is carried out with Frobenius norm. The accuracy obtained using our model was 0.41, and the top-5 accuracy obtained was 0.87.
Cite this Research Publication : Jain, Gokul S., H. B. Ganesh, N. S. Kamal, V. V. Variyar, V. Sowmya, and K. P. Soman. "Geometrical Feature Extraction of CAD Models with Fully Convolutional Networks." In Advanced Machine Intelligence and Signal Processing, pp. 845-853. Springer, Singapore, 2022.