Research on a 3D Point Cloud Map Learning Algorithm Based on Point Normal Constraints

Area of Science:

• Computer Vision
• Geometric Processing
• Signal Processing

Background:

• Laser point clouds are susceptible to Gaussian and Laplace noise, degrading surface reconstruction and visualization accuracy.
• Existing denoising methods often fail to consider local consistency and density of point cloud normal vectors.

Purpose of the Study:

• To develop an advanced point cloud denoising algorithm that addresses limitations of current methods.
• To enhance the accuracy, robustness, and efficiency of laser point cloud denoising.

Main Methods:

• A feature map learning algorithm integrating point normal constraints, Dirichlet energy, and coupled orthogonality bias terms.
• Utilizing Dirichlet energy to penalize differences between neighboring normal vectors.
• Incorporating a point cloud density function to capture local feature correlations and mitigate mixed noise.

Main Results:

• Reduced average Mean Square Error (MSE) by 0.005 and 0.054 compared to MRPCA and NLD algorithms.
• Improved average Signal-to-Noise Ratio (SNR) by 0.13 dB and 2.14 dB compared to MRPCA and AWLOP.
• Achieved a 27% increase in computational efficiency compared to the RSLDM method.

Conclusions:

• The proposed algorithm effectively removes mixed noise from laser point clouds while preserving essential local geometric features.
• The method demonstrates superior performance in accuracy, robustness, and computational efficiency over existing techniques.
• This approach offers a significant advancement for applications relying on high-quality point cloud data.