研究目的
To direct extract the feature descriptors of the point cloud object through the raw point cloud using a novel network based on the Siamese Network, PointNet, and PointNet++.
研究成果
The proposed network, based on the Siamese Network, effectively extracts feature descriptors from raw point clouds and demonstrates robust generalization across different datasets. The network's performance is validated through experiments on MLS and ModelNet40 datasets, showing good accuracy in feature descriptor extraction and classification.
研究不足
The study does not explicitly mention limitations, but potential areas for optimization could include handling more diverse point cloud objects and improving the network's performance on unseen data categories.
1：Experimental Design and Method Selection:
The study proposes a novel network framework based on the Siamese Network to directly extract feature descriptors from raw point clouds. The network includes a Feature Encode Network (FEN) and a discriminator network, with shared weights between the two branches of the Siamese network. The Euclidean distance is used as the loss function to reflect feature descriptors similarity.
2：Sample Selection and Data Sources:
The experiment uses datasets acquired by a Mobile Laser Scanning (MLS) system, containing 6 categories of point cloud objects. Additionally, the ModelNet40 benchmark dataset is used for training, which includes 12,311 CAD models from 40 man-made object categories.
3：List of Experimental Equipment and Materials:
The study utilizes a RIGEL VMX-450 system for data acquisition and a Titan X GPU for training the network.
4：Experimental Procedures and Operational Workflow:
The network is trained using the Adaptive Moment Estimation (Adam) Optimizer with a learning rate of 0.001. The training process stops when the loss function converges. The network's performance is tested on point cloud objects from both the MLS system and the ModelNet40 dataset.
5：The training process stops when the loss function converges. The network's performance is tested on point cloud objects from both the MLS system and the ModelNet40 dataset.
Data Analysis Methods:
5. Data Analysis Methods: The accuracy of the network is evaluated based on the search accuracy of real-valued descriptors and binary descriptors. The feature descriptors' consistency is visualized through histograms.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
