A robust cloud registration method based on redundant data reduction using backpropagation neural network and shift...

Area of Science:

• Computer Vision
• Computational Geometry
• Machine Learning

Background:

• Accurate registration of point clouds is crucial for 3D data processing.
• Existing methods can be computationally intensive, especially for large datasets.
• Efficient registration algorithms are needed to reduce processing time and resource requirements.

Purpose of the Study:

• To propose a robust and efficient coarse-to-fine registration method for point clouds.
• To leverage backpropagation (BP) neural networks and shift window technology for improved registration.
• To reduce the computational complexity of point cloud registration while maintaining accuracy.

Main Methods:

• Coarse alignment using initial rotation and translation estimation.
• Data simplification via BP neural network with contour point reservation.
• Fine registration employing the reweighted iterative closest point algorithm.

Main Results:

• Significant reduction in the number of points through data simplification.
• Substantial decrease in time and space complexity for accurate registration.
• Experimental validation demonstrating improved computational efficiency.

Conclusions:

• The proposed method effectively enhances computational efficiency in point cloud registration.
• The integration of BP neural networks and shift window technology offers a promising approach.
• The method achieves registration improvements without sacrificing accuracy.