Details

Autor(en) / Beteiligte

• Chang, Wen-Chung
• Pham, Van-Toan

Titel

An Efficient Neural Network with Performance-Based Switching of Candidate Optimizers for Point Cloud Matching

Ist Teil von

• Proceedings of the 6th International Conference on Control, Mechatronics and Automation, 2018, p.159-164

Ort / Verlag

New York, NY, USA: ACM

Erscheinungsjahr

2018

Quelle

ACM Digital Library Complete

Beschreibungen/Notizen

• Typically, Iterative Closet Point approach can be applied to perform point cloud matching tasks as long as the variation between two point clouds is not large. In order to perform such matching tasks robustly and efficiently, an effective neural network structure is presented in the paper. Particularly, the paper resolves the problem of improving convergence rate in training a neural network to estimate a rotation angle between two 2D point clouds. Firstly, in order to shorten the learning process of the neural network, an innovative parameter updating algorithm based on performance of different candidate optimizers is proposed. Secondly, the neural network is trained based on this algorithm to learn from point cloud data sets. Then, the trained neural network can determine the rotation angle between two 2D point clouds effectively. Finally, the performance of the approach developed in this paper is validated by experiments.